id
int64
1
42.8k
arxiv_id
stringlengths
10
13
base_arxiv_id
float64
1.11k
2.51k
version
float64
1
14
title
stringlengths
8
224
abstract
stringlengths
51
4.03k
submit_date
stringdate
2011-09-01 00:00:00
2025-08-13 00:00:00
metadata
stringlengths
283
56.2k
199
2508.00619
2,508.00619
null
DACTYL: Diverse Adversarial Corpus of Texts Yielded from Large Language Models
Existing AIG (AI-generated) text detectors struggle in real-world settings despite succeeding in internal testing, suggesting that they may not be robust enough. We rigorously examine the machine-learning procedure to build these detectors to address this. Most current AIG text detection datasets focus on zero-shot generations, but little work has been done on few-shot or one-shot generations, where LLMs are given human texts as an example. In response, we introduce the Diverse Adversarial Corpus of Texts Yielded from Language models (DACTYL), a challenging AIG text detection dataset focusing on one-shot/few-shot generations. We also include texts from domain-specific continued-pre-trained (CPT) language models, where we fully train all parameters using a memory-efficient optimization approach. Many existing AIG text detectors struggle significantly on our dataset, indicating a potential vulnerability to one-shot/few-shot and CPT-generated texts. We also train our own classifiers using two approaches: standard binary cross-entropy (BCE) optimization and a more recent approach, deep X-risk optimization (DXO). While BCE-trained classifiers marginally outperform DXO classifiers on the DACTYL test set, the latter excels on out-of-distribution (OOD) texts. In our mock deployment scenario in student essay detection with an OOD student essay dataset, the best DXO classifier outscored the best BCE-trained classifier by 50.56 macro-F1 score points at the lowest false positive rates for both. Our results indicate that DXO classifiers generalize better without overfitting to the test set. Our experiments highlight several areas of improvement for AIG text detectors.
2025-08-01
{'id': '2508.00619', 'title': 'DACTYL: Diverse Adversarial Corpus of Texts Yielded from Large Language Models', 'abstract': 'Existing AIG (AI-generated) text detectors struggle in real-world settings\ndespite succeeding in internal testing, suggesting that they may not be robust\nenough. We rigorously examine the machine-learning procedure to build these\ndetectors to address this. Most current AIG text detection datasets focus on\nzero-shot generations, but little work has been done on few-shot or one-shot\ngenerations, where LLMs are given human texts as an example. In response, we\nintroduce the Diverse Adversarial Corpus of Texts Yielded from Language models\n(DACTYL), a challenging AIG text detection dataset focusing on\none-shot/few-shot generations. We also include texts from domain-specific\ncontinued-pre-trained (CPT) language models, where we fully train all\nparameters using a memory-efficient optimization approach. Many existing AIG\ntext detectors struggle significantly on our dataset, indicating a potential\nvulnerability to one-shot/few-shot and CPT-generated texts. We also train our\nown classifiers using two approaches: standard binary cross-entropy (BCE)\noptimization and a more recent approach, deep X-risk optimization (DXO). While\nBCE-trained classifiers marginally outperform DXO classifiers on the DACTYL\ntest set, the latter excels on out-of-distribution (OOD) texts. In our mock\ndeployment scenario in student essay detection with an OOD student essay\ndataset, the best DXO classifier outscored the best BCE-trained classifier by\n50.56 macro-F1 score points at the lowest false positive rates for both. Our\nresults indicate that DXO classifiers generalize better without overfitting to\nthe test set. Our experiments highlight several areas of improvement for AIG\ntext detectors.', 'authors': ['Shantanu Thorat', 'Andrew Caines'], 'published': '2025-08-01 13:28:01+00:00', 'categories': ['cs.CL', 'cs.LG'], 'url': 'http://arxiv.org/abs/2508.00619v1'}
200
2508.00301
2,508.00301
null
Entangling Power and Its Deviation: A Quantitative Analysis on Input-State Dependence and Variability in Entanglement Generation
Quantifying the entangling capability of quantum operations is a fundamental task in quantum information science. Traditionally, this capability is measured by the entangling power (EP), defined as the average entanglement generated when a quantum operation acts uniformly on all possible product states. However, EP alone cannot capture the intricate input-state-dependent nature of entanglement generation. To address this, we define a complementary metric -- entangling power deviation (EPD) -- as the standard deviation of entanglement generated over all product input states, thereby capturing the multifaceted nature of entangling behavior. We develop a general group-theoretical framework that yields closed-form expressions for both EP and EPD. Our analysis reveals a fundamental and previously unexplored physics: enhancing entangling capability inevitably increases sensitivity, or bias, toward specific input states. By analyzing representative two-qubit gates, we show that the gates with identical EP can exhibit markedly different EPD values, illustrating that the nature of entanglement generation can significantly differ depending on physical implementation. Extending our framework to a class of generalized controlled-unitary operations acting on bipartite Hilbert spaces of arbitrary dimensions, we (re)affirm the inherent trade-off between the entangling strength and uniformity. Moreover, we uncover a subtle dimension-parity-dependent behavior in entanglement generation, which EP alone fails to detect. These findings highlight EPD as an indispensable diagnostic tool -- one that, alongside EP, provides a deeper and more complete characterization of the entangling structure.
2025-08-01
{'id': '2508.00301', 'title': 'Entangling Power and Its Deviation: A Quantitative Analysis on Input-State Dependence and Variability in Entanglement Generation', 'abstract': 'Quantifying the entangling capability of quantum operations is a fundamental\ntask in quantum information science. Traditionally, this capability is measured\nby the entangling power (EP), defined as the average entanglement generated\nwhen a quantum operation acts uniformly on all possible product states.\nHowever, EP alone cannot capture the intricate input-state-dependent nature of\nentanglement generation. To address this, we define a complementary metric --\nentangling power deviation (EPD) -- as the standard deviation of entanglement\ngenerated over all product input states, thereby capturing the multifaceted\nnature of entangling behavior. We develop a general group-theoretical framework\nthat yields closed-form expressions for both EP and EPD. Our analysis reveals a\nfundamental and previously unexplored physics: enhancing entangling capability\ninevitably increases sensitivity, or bias, toward specific input states. By\nanalyzing representative two-qubit gates, we show that the gates with identical\nEP can exhibit markedly different EPD values, illustrating that the nature of\nentanglement generation can significantly differ depending on physical\nimplementation. Extending our framework to a class of generalized\ncontrolled-unitary operations acting on bipartite Hilbert spaces of arbitrary\ndimensions, we (re)affirm the inherent trade-off between the entangling\nstrength and uniformity. Moreover, we uncover a subtle\ndimension-parity-dependent behavior in entanglement generation, which EP alone\nfails to detect. These findings highlight EPD as an indispensable diagnostic\ntool -- one that, alongside EP, provides a deeper and more complete\ncharacterization of the entangling structure.', 'authors': ['Kyoungho Cho', 'Jeongho Bang'], 'published': '2025-08-01 04:06:55+00:00', 'categories': ['quant-ph'], 'url': 'http://arxiv.org/abs/2508.00301v1'}
201
2508.00310
2,508.0031
null
A compact quasi-zero stiffness metamaterial based on monolithic shells for vibration isolation
Quasi-zero stiffness (QZS) metamaterials are highly effective in isolating objects from low-frequency external vibrations, due to their high static stiffness but low dynamic stiffness characteristics. Traditionally, QZS metamaterials are designed by combining a negative-stiffness part with a positive-stiffness counterpart. Here, we present a novel QZS metamaterial design without relying on combining two components. The QZS characteristic is achieved solely through monolithic shell elements' unique geometry and nonlinear deformation. Using experimental and numerical approaches, we investigate the static and dynamic responses of the proposed metamaterials as a function of their geometric parameters. We then tune the structure's geometry to achieve ideal zero-stiffness behaviors and experimentally demonstrate an exceptional low-frequency vibration isolation mechanism. This concept can be further utilized as a building block for constructing metamaterials with multiple zero-stiffness features, enabling a broad range of applications.
2025-08-01
{'id': '2508.00310', 'title': 'A compact quasi-zero stiffness metamaterial based on monolithic shells for vibration isolation', 'abstract': "Quasi-zero stiffness (QZS) metamaterials are highly effective in isolating\nobjects from low-frequency external vibrations, due to their high static\nstiffness but low dynamic stiffness characteristics. Traditionally, QZS\nmetamaterials are designed by combining a negative-stiffness part with a\npositive-stiffness counterpart. Here, we present a novel QZS metamaterial\ndesign without relying on combining two components. The QZS characteristic is\nachieved solely through monolithic shell elements' unique geometry and\nnonlinear deformation. Using experimental and numerical approaches, we\ninvestigate the static and dynamic responses of the proposed metamaterials as a\nfunction of their geometric parameters. We then tune the structure's geometry\nto achieve ideal zero-stiffness behaviors and experimentally demonstrate an\nexceptional low-frequency vibration isolation mechanism. This concept can be\nfurther utilized as a building block for constructing metamaterials with\nmultiple zero-stiffness features, enabling a broad range of applications.", 'authors': ['Yong Zhang', 'Xianfeng Chen'], 'published': '2025-08-01 04:32:12+00:00', 'categories': ['physics.app-ph'], 'url': 'http://arxiv.org/abs/2508.00310v1'}
202
2508.00353
2,508.00353
null
Nonclassical microwave radiation from the parametric dynamical Casimir effect in the reversed-dissipation regime of circuit optomechanics
We propose an experimentally feasible optomechanical system (OMS) that is dispersively driven and operates in the reversed dissipation regime (RDR), where the mechanical damping rate far exceeds the cavity decay rate. We demonstrate that coherent, fast-time modulation of the driving laser frequency-on time scales longer than the mechanical decoherence time-allows for adiabatic elimination of the mechanical mode, resulting in strong parametric amplification of quantum vacuum fluctuations of the intracavity field. This mechanism, known as the parametric dynamical Casimir effect (parametric-DCE), leads to the generation of Casimir photons. In the dispersive RDR, we find that the total system Hamiltonian-including the DCE term-is intrinsically modified by a generalized optomechanical Kerr-type nonlinearity. This nonlinearity not only saturates the mean number of radiated Casimir photons on short time scales, even without dissipation, but also induces oscillatory behavior in their dynamics and quantum characteristics. Remarkably, the presence of the Kerr nonlinearity causes the generated DCE photons to exhibit nonclassical features, including sub-Poissonian statistics, negative Wigner function and quadrature squeezing which can be controlled by adjusting the system parameters. The proposed nonclassical microwave radiation source possesses the potential to be applied in quantum information processing, quantum computing as well as microwave quantum sensing.
2025-08-01
{'id': '2508.00353', 'title': 'Nonclassical microwave radiation from the parametric dynamical Casimir effect in the reversed-dissipation regime of circuit optomechanics', 'abstract': 'We propose an experimentally feasible optomechanical system (OMS) that is\ndispersively driven and operates in the reversed dissipation regime (RDR),\nwhere the mechanical damping rate far exceeds the cavity decay rate. We\ndemonstrate that coherent, fast-time modulation of the driving laser\nfrequency-on time scales longer than the mechanical decoherence time-allows for\nadiabatic elimination of the mechanical mode, resulting in strong parametric\namplification of quantum vacuum fluctuations of the intracavity field. This\nmechanism, known as the parametric dynamical Casimir effect (parametric-DCE),\nleads to the generation of Casimir photons. In the dispersive RDR, we find that\nthe total system Hamiltonian-including the DCE term-is intrinsically modified\nby a generalized optomechanical Kerr-type nonlinearity. This nonlinearity not\nonly saturates the mean number of radiated Casimir photons on short time\nscales, even without dissipation, but also induces oscillatory behavior in\ntheir dynamics and quantum characteristics. Remarkably, the presence of the\nKerr nonlinearity causes the generated DCE photons to exhibit nonclassical\nfeatures, including sub-Poissonian statistics, negative Wigner function and\nquadrature squeezing which can be controlled by adjusting the system\nparameters. The proposed nonclassical microwave radiation source possesses the\npotential to be applied in quantum information processing, quantum computing as\nwell as microwave quantum sensing.', 'authors': ['H. Solki', 'Ali Motazedifard', 'M. H. Naderi', 'A. Youssefi', 'R. Roknizadeh'], 'published': '2025-08-01 06:36:05+00:00', 'categories': ['quant-ph'], 'url': 'http://arxiv.org/abs/2508.00353v1'}
203
2508.00350
2,508.0035
null
BOOD: Boundary-based Out-Of-Distribution Data Generation
Harnessing the power of diffusion models to synthesize auxiliary training data based on latent space features has proven effective in enhancing out-of-distribution (OOD) detection performance. However, extracting effective features outside the in-distribution (ID) boundary in latent space remains challenging due to the difficulty of identifying decision boundaries between classes. This paper proposes a novel framework called Boundary-based Out-Of-Distribution data generation (BOOD), which synthesizes high-quality OOD features and generates human-compatible outlier images using diffusion models. BOOD first learns a text-conditioned latent feature space from the ID dataset, selects ID features closest to the decision boundary, and perturbs them to cross the decision boundary to form OOD features. These synthetic OOD features are then decoded into images in pixel space by a diffusion model. Compared to previous works, BOOD provides a more training efficient strategy for synthesizing informative OOD features, facilitating clearer distinctions between ID and OOD data. Extensive experimental results on common benchmarks demonstrate that BOOD surpasses the state-of-the-art method significantly, achieving a 29.64% decrease in average FPR95 (40.31% vs. 10.67%) and a 7.27% improvement in average AUROC (90.15% vs. 97.42%) on the CIFAR-100 dataset.
2025-08-01
{'id': '2508.00350', 'title': 'BOOD: Boundary-based Out-Of-Distribution Data Generation', 'abstract': 'Harnessing the power of diffusion models to synthesize auxiliary training\ndata based on latent space features has proven effective in enhancing\nout-of-distribution (OOD) detection performance. However, extracting effective\nfeatures outside the in-distribution (ID) boundary in latent space remains\nchallenging due to the difficulty of identifying decision boundaries between\nclasses. This paper proposes a novel framework called Boundary-based\nOut-Of-Distribution data generation (BOOD), which synthesizes high-quality OOD\nfeatures and generates human-compatible outlier images using diffusion models.\nBOOD first learns a text-conditioned latent feature space from the ID dataset,\nselects ID features closest to the decision boundary, and perturbs them to\ncross the decision boundary to form OOD features. These synthetic OOD features\nare then decoded into images in pixel space by a diffusion model. Compared to\nprevious works, BOOD provides a more training efficient strategy for\nsynthesizing informative OOD features, facilitating clearer distinctions\nbetween ID and OOD data. Extensive experimental results on common benchmarks\ndemonstrate that BOOD surpasses the state-of-the-art method significantly,\nachieving a 29.64% decrease in average FPR95 (40.31% vs. 10.67%) and a 7.27%\nimprovement in average AUROC (90.15% vs. 97.42%) on the CIFAR-100 dataset.', 'authors': ['Qilin Liao', 'Shuo Yang', 'Bo Zhao', 'Ping Luo', 'Hengshuang Zhao'], 'published': '2025-08-01 06:34:27+00:00', 'categories': ['cs.LG'], 'url': 'http://arxiv.org/abs/2508.00350v1'}
204
2508.00252
2,508.00252
null
TofuML: A Spatio-Physical Interactive Machine Learning Device for Interactive Exploration of Machine Learning for Novices
We introduce TofuML, an interactive system designed to make machine learning (ML) concepts more accessible and engaging for non-expert users. Unlike conventional GUI-based systems, TofuML employs a physical and spatial interface consisting of a small device and a paper mat, allowing users to train and evaluate sound classification models through intuitive, toy-like interactions. Through two user studies -- a comparative study against a GUI-based version and a public event deployment -- we investigated how TofuML impacts users' engagement in the ML model creation process, their ability to provide appropriate training data, and their conception of potential applications. Our results indicated that TofuML enhanced user engagement compared to a GUI while lowering barriers for non-experts to engage with ML. Users demonstrated creativity in conceiving diverse ML applications, revealing opportunities to optimize between conceptual understanding and user engagement. These findings contribute to developing interactive ML systems/frameworks designed for a wide range of users.
2025-08-01
{'id': '2508.00252', 'title': 'TofuML: A Spatio-Physical Interactive Machine Learning Device for Interactive Exploration of Machine Learning for Novices', 'abstract': "We introduce TofuML, an interactive system designed to make machine learning\n(ML) concepts more accessible and engaging for non-expert users. Unlike\nconventional GUI-based systems, TofuML employs a physical and spatial interface\nconsisting of a small device and a paper mat, allowing users to train and\nevaluate sound classification models through intuitive, toy-like interactions.\nThrough two user studies -- a comparative study against a GUI-based version and\na public event deployment -- we investigated how TofuML impacts users'\nengagement in the ML model creation process, their ability to provide\nappropriate training data, and their conception of potential applications. Our\nresults indicated that TofuML enhanced user engagement compared to a GUI while\nlowering barriers for non-experts to engage with ML. Users demonstrated\ncreativity in conceiving diverse ML applications, revealing opportunities to\noptimize between conceptual understanding and user engagement. These findings\ncontribute to developing interactive ML systems/frameworks designed for a wide\nrange of users.", 'authors': ['Wataru Kawabe', 'Hiroto Fukuda', 'Akihisa Shitara', 'Yuri Nakao', 'Yusuke Sugano'], 'published': '2025-08-01 01:44:56+00:00', 'categories': ['cs.HC'], 'url': 'http://arxiv.org/abs/2508.00252v2'}
205
2508.00559
2,508.00559
null
Solitary-wave solutions of the fractional nonlinear Schrödinger equation. II. A numerical study of the dynamics
The present paper is a numerical study of the dynamics of solitary wave solutions of the fractional nonlinear Schr\"odinger equation, whose existence was analyzed by the authors in the first part of the project. The computational study will be made from the approximation of the periodic initial-value problem with a fully discrete scheme consisting of a Fourier spectral method for the spatial discretization and a fourth-order, Runge-Kutta-Composition method as time integrator. Several issues regarding the stability of the waves, such as the effects of small and large perturbations, interactions of solitary waves and the resolution of initial data into trains of waves are discussed.
2025-08-01
{'id': '2508.00559', 'title': 'Solitary-wave solutions of the fractional nonlinear Schrödinger equation. II. A numerical study of the dynamics', 'abstract': 'The present paper is a numerical study of the dynamics of solitary wave\nsolutions of the fractional nonlinear Schr\\"{o}dinger equation, whose existence\nwas analyzed by the authors in the first part of the project. The computational\nstudy will be made from the approximation of the periodic initial-value problem\nwith a fully discrete scheme consisting of a Fourier spectral method for the\nspatial discretization and a fourth-order, Runge-Kutta-Composition method as\ntime integrator. Several issues regarding the stability of the waves, such as\nthe effects of small and large perturbations, interactions of solitary waves\nand the resolution of initial data into trains of waves are discussed.', 'authors': ['Angel Durán', 'Nuria Reguera'], 'published': '2025-08-01 11:56:38+00:00', 'categories': ['math.NA', 'cs.NA', 'math.AP', '76B25, 35C07, 65H10'], 'url': 'http://arxiv.org/abs/2508.00559v1'}
206
2508.00470
2,508.0047
null
Hyperon Pair Production at BESIII
This poster presents recent experimental findings from the BESIII experiment, including two categories of studies. Firstly, it reports the studies on the hyperon pair production in the processes $e^+e^- $, $^+^-$, $^0^0$, $^0^0$ and $^-^+$. The Born cross sections for these processes are measured at center-of-mass energies ranging from $3.5$ to $4.9$ GeV. For the first time, evidence for the decays $(3770)$ and $^-^+$ is found. Secondly, it reports the measurements of the Born cross sections of multi-body decays involving hyperon final states. The evidence of $(4160)K^-^++c.c.$ is found. These results offer new perspectives on hyperon production in $e^+e^-$ annihilation and the decay mechanism of vector charmonium(-like) states, contributing to our understanding of hadron dynamics.
2025-08-01
{'id': '2508.00470', 'title': 'Hyperon Pair Production at BESIII', 'abstract': 'This poster presents recent experimental findings from the BESIII experiment,\nincluding two categories of studies. Firstly, it reports the studies on the\nhyperon pair production in the processes $e^+e^- \\to \\Lambda\\bar\\Lambda$,\n$\\Sigma^+\\bar\\Sigma^-$, $\\Sigma^0\\bar\\Sigma^0$, $\\Xi^0\\bar\\Xi^0$ and\n$\\Xi^-\\bar\\Xi^+$. The Born cross sections for these processes are measured at\ncenter-of-mass energies ranging from $3.5$ to $4.9$ GeV. For the first time,\nevidence for the decays $\\psi(3770)\\to\\Lambda\\bar\\Lambda$ and $\\Xi^-\\bar\\Xi^+$\nis found. Secondly, it reports the measurements of the Born cross sections of\nmulti-body decays involving hyperon final states. The evidence of\n$\\psi(4160)\\to K^-\\bar\\Xi^+\\Lambda+c.c.$ is found. These results offer new\nperspectives on hyperon production in $e^+e^-$ annihilation and the decay\nmechanism of vector charmonium(-like) states, contributing to our understanding\nof hadron dynamics.', 'authors': ['Zijie Shang', 'Ruoyu Zhang', 'Xiongfei wang'], 'published': '2025-08-01 09:45:15+00:00', 'categories': ['hep-ex'], 'url': 'http://arxiv.org/abs/2508.00470v1'}
207
2508.00294
2,508.00294
null
Formal Power Series Representations in Probability and Expected Utility Theory
We advance a general theory of coherent preference that surrenders restrictions embodied in orthodox doctrine. This theory enjoys the property that any preference system admits extension to a complete system of preferences, provided it satisfies a certain coherence requirement analogous to the one de Finetti advanced for his foundations of probability. Unlike de Finetti's theory, the one we set forth requires neither transitivity nor Archimedeanness nor boundedness nor continuity of preference. This theory also enjoys the property that any complete preference system meeting the standard of coherence can be represented by utility in an ordered field extension of the reals. Representability by utility is a corollary of this paper's central result, which at once extends H\"older's Theorem and strengthens Hahn's Embedding Theorem.
2025-08-01
{'id': '2508.00294', 'title': 'Formal Power Series Representations in Probability and Expected Utility Theory', 'abstract': 'We advance a general theory of coherent preference that surrenders\nrestrictions embodied in orthodox doctrine. This theory enjoys the property\nthat any preference system admits extension to a complete system of\npreferences, provided it satisfies a certain coherence requirement analogous to\nthe one de Finetti advanced for his foundations of probability. Unlike de\nFinetti\'s theory, the one we set forth requires neither transitivity nor\nArchimedeanness nor boundedness nor continuity of preference. This theory also\nenjoys the property that any complete preference system meeting the standard of\ncoherence can be represented by utility in an ordered field extension of the\nreals. Representability by utility is a corollary of this paper\'s central\nresult, which at once extends H\\"older\'s Theorem and strengthens Hahn\'s\nEmbedding Theorem.', 'authors': ['Arthur Paul Pedersen', 'Samuel Allen Alexander'], 'published': '2025-08-01 03:34:39+00:00', 'categories': ['math.PR', 'cs.AI', 'econ.TH', 'math.LO', 'math.ST', 'stat.TH', '60A05'], 'url': 'http://arxiv.org/abs/2508.00294v1'}
208
2508.00327
2,508.00327
null
Etching-to-deposition transition in SiO$_2$/Si$_3$N$_4$ using CH$_x$F$_y$ ion-based plasma etching: An atomistic study with neural network potentials
Plasma etching, a critical process in semiconductor fabrication, utilizes hydrofluorocarbons both as etchants and as precursors for carbon film formation, where precise control over film growth is essential for achieving high SiO$_2$/Si$_3$N$_4$ selectivity and enabling atomic layer etching. In this work, we develop neural network potentials (NNPs) to gain atomistic insights into the surface evolution of SiO$_2$ and Si$_3$N$_4$ under hydrofluorocarbon ion bombardment. To efficiently sample diverse local configurations without exhaustive enumeration of ion-substrate combinations, we propose a vapor-to-surface sampling approach using high-temperature, low-density molecular dynamics simulations, supplemented with baseline reference structures. The NNPs, refined through iterative training, yield etching characteristics in MD simulations that show good agreement with experimental results. Further analysis reveals distinct mechanisms of carbon layer formation in SiO$_2$ and Si$_3$N$_4$, driven by the higher volatility of carbon-oxygen byproducts in SiO$_2$ and the suppressed formation of volatile carbon-nitrogen species in Si$_3$N$_4$. This computational framework enables quantitative predictions of atomistic surface modifications under plasma exposure and provides a foundation for integration with multiscale process modeling, offering insights into semiconductor fabrication processes.
2025-08-01
{'id': '2508.00327', 'title': 'Etching-to-deposition transition in SiO$_2$/Si$_3$N$_4$ using CH$_x$F$_y$ ion-based plasma etching: An atomistic study with neural network potentials', 'abstract': 'Plasma etching, a critical process in semiconductor fabrication, utilizes\nhydrofluorocarbons both as etchants and as precursors for carbon film\nformation, where precise control over film growth is essential for achieving\nhigh SiO$_2$/Si$_3$N$_4$ selectivity and enabling atomic layer etching. In this\nwork, we develop neural network potentials (NNPs) to gain atomistic insights\ninto the surface evolution of SiO$_2$ and Si$_3$N$_4$ under hydrofluorocarbon\nion bombardment. To efficiently sample diverse local configurations without\nexhaustive enumeration of ion-substrate combinations, we propose a\nvapor-to-surface sampling approach using high-temperature, low-density\nmolecular dynamics simulations, supplemented with baseline reference\nstructures. The NNPs, refined through iterative training, yield etching\ncharacteristics in MD simulations that show good agreement with experimental\nresults. Further analysis reveals distinct mechanisms of carbon layer formation\nin SiO$_2$ and Si$_3$N$_4$, driven by the higher volatility of carbon-oxygen\nbyproducts in SiO$_2$ and the suppressed formation of volatile carbon-nitrogen\nspecies in Si$_3$N$_4$. This computational framework enables quantitative\npredictions of atomistic surface modifications under plasma exposure and\nprovides a foundation for integration with multiscale process modeling,\noffering insights into semiconductor fabrication processes.', 'authors': ['Hyungmin An', 'Sangmin Oh', 'Dongheon Lee', 'Jae-hyeon Ko', 'Dongyean Oh', 'Changho Hong', 'Seungwu Han'], 'published': '2025-08-01 05:20:45+00:00', 'categories': ['cond-mat.mtrl-sci'], 'url': 'http://arxiv.org/abs/2508.00327v1'}
209
2508.00262
2,508.00262
null
Towards Efficient Verification of Computation in Quantum Devices
Designing quantum processors is a complex task that demands advanced verification methods to ensure their correct functionality. However, traditional methods of comprehensively verifying quantum devices, such as quantum process tomography, face significant limitations because of the exponential growth in computational resources. These limitations arise from treating the system as a black box and ignoring its design structure. Consequently, new testing methods must be developed considering the design structure. In this paper, we investigate the structure of computations on the hardware, focusing on the layered interruptible quantum circuit model and designing a scalable algorithm to verify it comprehensively. Specifically, for a given quantum hardware that claims to process an unknown $n$ qubit $d$ layer circuit via a finite set of quantum gates, our method completely reconstructs the circuits within a time complexity of $O(d^2 t (n/))$, guaranteeing success with a probability of at least $1-$. Here, $t$ represents the maximum execution time for each circuit layer. Our approach significantly reduces execution time for completely verifying computations in quantum devices, achieving double logarithmic scaling in the problem size. Furthermore, we validate our algorithm through experiments using IBM's quantum cloud service, demonstrating its potential applicability in the noisy intermediate-scale quantum era.
2025-08-01
{'id': '2508.00262', 'title': 'Towards Efficient Verification of Computation in Quantum Devices', 'abstract': "Designing quantum processors is a complex task that demands advanced\nverification methods to ensure their correct functionality. However,\ntraditional methods of comprehensively verifying quantum devices, such as\nquantum process tomography, face significant limitations because of the\nexponential growth in computational resources. These limitations arise from\ntreating the system as a black box and ignoring its design structure.\nConsequently, new testing methods must be developed considering the design\nstructure. In this paper, we investigate the structure of computations on the\nhardware, focusing on the layered interruptible quantum circuit model and\ndesigning a scalable algorithm to verify it comprehensively. Specifically, for\na given quantum hardware that claims to process an unknown $n$ qubit $d$ layer\ncircuit via a finite set of quantum gates, our method completely reconstructs\nthe circuits within a time complexity of $O(d^2 t \\log (n/\\delta))$,\nguaranteeing success with a probability of at least $1-\\delta$. Here, $t$\nrepresents the maximum execution time for each circuit layer. Our approach\nsignificantly reduces execution time for completely verifying computations in\nquantum devices, achieving double logarithmic scaling in the problem size.\nFurthermore, we validate our algorithm through experiments using IBM's quantum\ncloud service, demonstrating its potential applicability in the noisy\nintermediate-scale quantum era.", 'authors': ['Keren Li', 'Peng Yan', 'Hanru Jiang', 'Nengkun Yu'], 'published': '2025-08-01 02:10:06+00:00', 'categories': ['quant-ph'], 'url': 'http://arxiv.org/abs/2508.00262v1'}
210
2508.00407
2,508.00407
null
Thermoelectric figure of merit and the deconfinement phase transition
Thermoelectric phenomena are traditionally associated with the interconversion of thermal and electrical energy in many-body systems. In the context of high-temperature quantum chromodynamics (QCD) matter produced in relativistic heavy-ion collisions, thermoelectric responses can provide insight into the evolving microscopic dynamics and the redistribution of effective degrees of freedom across the phase transition region. In this work, for the first time, we present a phenomenological study of the thermoelectric figure of merit (\( ZT \)) in hot QCD matter, with a particular focus on its behavior across the hadronic and quark-gluon plasma phases. Using model-based calculations for the electrical conductivity, Seebeck coefficient, and thermal conductivity, we analyze the temperature dependence of \( ZT \) and identify characteristic features near the QCD phase transition temperature. Our results indicate that \( ZT \) exhibits nontrivial behavior near the transition region, reflecting the changing transport properties and active degrees of freedom in the medium. This phenomenological study of the thermoelectric figure of merit provides a complementary perspective to traditional transport studies and may provide critical insights for advancing the understanding of QCD matter through the transition region.
2025-08-01
{'id': '2508.00407', 'title': 'Thermoelectric figure of merit and the deconfinement phase transition', 'abstract': 'Thermoelectric phenomena are traditionally associated with the\ninterconversion of thermal and electrical energy in many-body systems. In the\ncontext of high-temperature quantum chromodynamics (QCD) matter produced in\nrelativistic heavy-ion collisions, thermoelectric responses can provide insight\ninto the evolving microscopic dynamics and the redistribution of effective\ndegrees of freedom across the phase transition region. In this work, for the\nfirst time, we present a phenomenological study of the thermoelectric figure of\nmerit (\\( ZT \\)) in hot QCD matter, with a particular focus on its behavior\nacross the hadronic and quark-gluon plasma phases. Using model-based\ncalculations for the electrical conductivity, Seebeck coefficient, and thermal\nconductivity, we analyze the temperature dependence of \\( ZT \\) and identify\ncharacteristic features near the QCD phase transition temperature. Our results\nindicate that \\( ZT \\) exhibits nontrivial behavior near the transition region,\nreflecting the changing transport properties and active degrees of freedom in\nthe medium. This phenomenological study of the thermoelectric figure of merit\nprovides a complementary perspective to traditional transport studies and may\nprovide critical insights for advancing the understanding of QCD matter through\nthe transition region.', 'authors': ['Kamaljeet Singh', 'Raghunath Sahoo'], 'published': '2025-08-01 08:06:42+00:00', 'categories': ['hep-ph', 'hep-ex', 'hep-th', 'nucl-ex', 'nucl-th'], 'url': 'http://arxiv.org/abs/2508.00407v1'}
211
2508.00547
2,508.00547
null
Dirac operators for algebraic families
We introduce algebraic families of Dirac operators for the deformation family (and other related families) associated with a real reductive Lie group that interpolates the reductive group and the corresponding Cartan motion group. We prove Vogan's conjecture in this setting, relating the infinitesimal character of an algebraic family of Harish-Chandra modules and its Dirac cohomology.
2025-08-01
{'id': '2508.00547', 'title': 'Dirac operators for algebraic families', 'abstract': "We introduce algebraic families of Dirac operators for the deformation family\n(and other related families) associated with a real reductive Lie group that\ninterpolates the reductive group and the corresponding Cartan motion group. We\nprove Vogan's conjecture in this setting, relating the infinitesimal character\nof an algebraic family of Harish-Chandra modules and its Dirac cohomology.", 'authors': ['Spyridon Afentoulidis-Almpanis', 'Eyal Subag'], 'published': '2025-08-01 11:42:24+00:00', 'categories': ['math.RT', 'math.RA', '22E47 (Primary), 17B10, 17B20, 14D99 (Secondary)'], 'url': 'http://arxiv.org/abs/2508.00547v1'}
212
2508.00585
2,508.00585
null
A sample of giant radio sources from the NVGRC catalog
We present the results of a search for megaparsec-scale sources in the NVGRC catalog of candidates of giant radio source (GRS) based on the NVSS sky survey. We visually inspected 370 NVGRC sources, as well as radio sources falling within a neighborhood of about one square degree around the target object. In the studied sample, 48% of objects were classified as giant radio sources, 14% as sources with a projected linear size of less than 0.72 Mpc, and 38\% as physically unrelated objects combined by the recognition algorithm into one radio source. We identified 197 gaints, of which 72 radio sources are known GRGs or GRQs, and 125 sources were identified by us as GRS for the first time. Comparing the proportion of FRI giants in four redshift bins, we found that for z<0.05, the proportions of FRI and FRII sources were approximately equal, but already at z>0.15 the proportion of FRI giants decreases sharply. The predominance of FRII giants in the GRS lists is most likely due to observational selection due to the sensitivity limit of existing radio surveys. Comparing the NVSS and VLASS cutouts, we found that 33% of sources can be classified as fadded. 25% of the sources show a restart of the radio source phase. 38% of the sources have deformed radio lobes. Our GRS sample includes 74% of galaxies, 15% of IR-excess galaxies, which, according to the WISE photometric data, can be attributed to quasars, and 11% of quasars. When visually examining the optical survey cutouts, we noted the presence of close neighbors for the hosts and/or their belonging to known groups or clusters of galaxies. Close neighbors at a distance of less than 50 kpc were found for 39% of radio sources, and 28% of sources are part of groups or clusters of galaxies. Thus, about 70% of gaints are in a fairly dense environment, and this proportion may be higher.
2025-08-01
{'id': '2508.00585', 'title': 'A sample of giant radio sources from the NVGRC catalog', 'abstract': 'We present the results of a search for megaparsec-scale sources in the NVGRC\ncatalog of candidates of giant radio source (GRS) based on the NVSS sky survey.\nWe visually inspected 370 NVGRC sources, as well as radio sources falling\nwithin a neighborhood of about one square degree around the target object. In\nthe studied sample, 48% of objects were classified as giant radio sources, 14%\nas sources with a projected linear size of less than 0.72 Mpc, and 38\\% as\nphysically unrelated objects combined by the recognition algorithm into one\nradio source. We identified 197 gaints, of which 72 radio sources are known\nGRGs or GRQs, and 125 sources were identified by us as GRS for the first time.\nComparing the proportion of FRI giants in four redshift bins, we found that for\nz<0.05, the proportions of FRI and FRII sources were approximately equal, but\nalready at z>0.15 the proportion of FRI giants decreases sharply. The\npredominance of FRII giants in the GRS lists is most likely due to\nobservational selection due to the sensitivity limit of existing radio surveys.\nComparing the NVSS and VLASS cutouts, we found that 33% of sources can be\nclassified as fadded. 25% of the sources show a restart of the radio source\nphase. 38% of the sources have deformed radio lobes. Our GRS sample includes\n74% of galaxies, 15% of IR-excess galaxies, which, according to the WISE\nphotometric data, can be attributed to quasars, and 11% of quasars. When\nvisually examining the optical survey cutouts, we noted the presence of close\nneighbors for the hosts and/or their belonging to known groups or clusters of\ngalaxies. Close neighbors at a distance of less than 50 kpc were found for 39%\nof radio sources, and 28% of sources are part of groups or clusters of\ngalaxies. Thus, about 70% of gaints are in a fairly dense environment, and this\nproportion may be higher.', 'authors': ['Olga Zhelenkova', 'Milena Khoruzhenko'], 'published': '2025-08-01 12:36:04+00:00', 'categories': ['astro-ph.GA', 'astro-ph.CO'], 'url': 'http://arxiv.org/abs/2508.00585v1'}
213
2508.00752
2,508.00752
null
The representation theory of somewhere-to-below shuffles
The *somewhere-to-below shuffles* are the elements \[ t_ := cyc_+cyc_,+1+cyc_,+1,+2++cyc_,+1,,n \] (for $\1,2,,n\$) in the group algebra $k[S_n]$ of the $n$-th symmetric group $S_n$. Their linear combinations are called the *one-sided cycle shuffles*. We determine the eigenvalues of the action of any one-sided cycle shuffle on any Specht module $S^$ of $S_n$.
2025-08-01
{'id': '2508.00752', 'title': 'The representation theory of somewhere-to-below shuffles', 'abstract': 'The *somewhere-to-below shuffles* are the elements \\[ t_{\\ell} :=\n\\operatorname{cyc}_{\\ell}+\\operatorname{cyc}_{\\ell,\\ell+1}+\\operatorname{cyc}_{\\ell,\\ell+1,\\ell+2}+\\cdots+\\operatorname{cyc}_{\\ell,\\ell+1,\\ldots,n}\n\\] (for $\\ell \\in \\{1,2,\\dots,n\\}$) in the group algebra $\\mathbf{k}[S_n]$ of\nthe $n$-th symmetric group $S_n$. Their linear combinations are called the\n*one-sided cycle shuffles*. We determine the eigenvalues of the action of any\none-sided cycle shuffle on any Specht module $\\mathcal{S}^{\\lambda}$ of $S_n$.', 'authors': ['Darij Grinberg'], 'published': '2025-08-01 16:29:40+00:00', 'categories': ['math.CO', 'math.RT', '05E99, 20C30, 60J10'], 'url': 'http://arxiv.org/abs/2508.00752v1'}
214
2508.00251
2,508.00251
null
Robust Model Reconstruction Based on the Topological Understanding of Point Clouds Using Persistent Homology
Reconstructing models from unorganized point clouds presents a significant challenge, especially when the models consist of multiple components represented by their surface point clouds. Such models often involve point clouds with noise that represent multiple closed surfaces with shared regions, making their automatic identification and separation inherently complex. In this paper, we propose an automatic method that uses the topological understanding provided by persistent homology, along with representative 2-cycles of persistent homology groups, to effectively distinguish and separate each closed surface. Furthermore, we employ Loop subdivision and least squares progressive iterative approximation (LSPIA) techniques to generate high-quality final surfaces and achieve complete model reconstruction. Our method is robust to noise in the point cloud, making it suitable for reconstructing models from such data. Experimental results demonstrate the effectiveness of our approach and highlight its potential for practical applications.
2025-08-01
{'id': '2508.00251', 'title': 'Robust Model Reconstruction Based on the Topological Understanding of Point Clouds Using Persistent Homology', 'abstract': 'Reconstructing models from unorganized point clouds presents a significant\nchallenge, especially when the models consist of multiple components\nrepresented by their surface point clouds. Such models often involve point\nclouds with noise that represent multiple closed surfaces with shared regions,\nmaking their automatic identification and separation inherently complex. In\nthis paper, we propose an automatic method that uses the topological\nunderstanding provided by persistent homology, along with representative\n2-cycles of persistent homology groups, to effectively distinguish and separate\neach closed surface. Furthermore, we employ Loop subdivision and least squares\nprogressive iterative approximation (LSPIA) techniques to generate high-quality\nfinal surfaces and achieve complete model reconstruction. Our method is robust\nto noise in the point cloud, making it suitable for reconstructing models from\nsuch data. Experimental results demonstrate the effectiveness of our approach\nand highlight its potential for practical applications.', 'authors': ['Yu Chen', 'Hongwei Lin'], 'published': '2025-08-01 01:43:06+00:00', 'categories': ['cs.CG'], 'url': 'http://arxiv.org/abs/2508.00251v1'}
216
2508.00258
2,508.00258
null
Topology-Inspired Morphological Descriptor for Soft Continuum Robots
This paper presents a topology-inspired morphological descriptor for soft continuum robots by combining a pseudo-rigid-body (PRB) model with Morse theory to achieve a quantitative characterization of robot morphologies. By counting critical points of directional projections, the proposed descriptor enables a discrete representation of multimodal configurations and facilitates morphological classification. Furthermore, we apply the descriptor to morphology control by formulating the target configuration as an optimization problem to compute actuation parameters that generate equilibrium shapes with desired topological features. The proposed framework provides a unified methodology for quantitative morphology description, classification, and control of soft continuum robots, with the potential to enhance their precision and adaptability in medical applications such as minimally invasive surgery and endovascular interventions.
2025-08-01
{'id': '2508.00258', 'title': 'Topology-Inspired Morphological Descriptor for Soft Continuum Robots', 'abstract': 'This paper presents a topology-inspired morphological descriptor for soft\ncontinuum robots by combining a pseudo-rigid-body (PRB) model with Morse theory\nto achieve a quantitative characterization of robot morphologies. By counting\ncritical points of directional projections, the proposed descriptor enables a\ndiscrete representation of multimodal configurations and facilitates\nmorphological classification. Furthermore, we apply the descriptor to\nmorphology control by formulating the target configuration as an optimization\nproblem to compute actuation parameters that generate equilibrium shapes with\ndesired topological features. The proposed framework provides a unified\nmethodology for quantitative morphology description, classification, and\ncontrol of soft continuum robots, with the potential to enhance their precision\nand adaptability in medical applications such as minimally invasive surgery and\nendovascular interventions.', 'authors': ['Zhiwei Wu', 'Siyi Wei', 'Jiahao Luo', 'Jinhui Zhang'], 'published': '2025-08-01 01:55:58+00:00', 'categories': ['cs.RO'], 'url': 'http://arxiv.org/abs/2508.00258v1'}
217
2508.00457
2,508.00457
null
Effects of inhomogenuity and anisotropy of radiation field on production and absorption of high energy radiation
We investigate the geometrical effects affecting the production and absorption of gamma-ray radiation emitted in inverse Compton scattering in the synchrotron-self-Compton process. We evaluate the effect of the anisotropy of the radiation field seen by the scattering electrons homogeneously distributed in the emission region. Next, we also consider inhomogeneous distribution of electrons and investigate the effect of it in the spherically symmetric emission region. We also study a cylindrical shape of the emission region and its effect on the isotropy of the emitted radiation. We obtain simple numerical factors that scale the emission for different assumptions about the geometry of the emission region and the distribution of the emitting electrons. For a 3D Gaussian spatial distribution of the electrons we obtain 0.222 times lower flux than for homogeneous emission region. Finally, we also evaluate the absorption of the radiation produced in the different scenarios, and compare the full calculations with the two most commonly assumed simplifications. We find that for cases when the absorption is lesser than by one order of magnitude, the full calculations for homogeneous sphere can be well approximated with homogeneously-emitting slab, while the absorption in the case of 3D Gaussian distribution of electrons is significantly weaker.
2025-08-01
{'id': '2508.00457', 'title': 'Effects of inhomogenuity and anisotropy of radiation field on production and absorption of high energy radiation', 'abstract': 'We investigate the geometrical effects affecting the production and\nabsorption of gamma-ray radiation emitted in inverse Compton scattering in the\nsynchrotron-self-Compton process. We evaluate the effect of the anisotropy of\nthe radiation field seen by the scattering electrons homogeneously distributed\nin the emission region. Next, we also consider inhomogeneous distribution of\nelectrons and investigate the effect of it in the spherically symmetric\nemission region. We also study a cylindrical shape of the emission region and\nits effect on the isotropy of the emitted radiation. We obtain simple numerical\nfactors that scale the emission for different assumptions about the geometry of\nthe emission region and the distribution of the emitting electrons. For a 3D\nGaussian spatial distribution of the electrons we obtain 0.222 times lower flux\nthan for homogeneous emission region. Finally, we also evaluate the absorption\nof the radiation produced in the different scenarios, and compare the full\ncalculations with the two most commonly assumed simplifications. We find that\nfor cases when the absorption is lesser than by one order of magnitude, the\nfull calculations for homogeneous sphere can be well approximated with\nhomogeneously-emitting slab, while the absorption in the case of 3D Gaussian\ndistribution of electrons is significantly weaker.', 'authors': ['J. Sitarek'], 'published': '2025-08-01 09:28:48+00:00', 'categories': ['astro-ph.HE'], 'url': 'http://arxiv.org/abs/2508.00457v1'}
218
2508.00582
2,508.00582
null
Diagnosing Floquet Chern and anomalous topological insulators based on Bloch oscillations
It is challenging to distinguish Floquet Chern insulator (FCI) and Floquet anomalous topological insulator (FATI) because of their common features of chiral edge states and far away from equilibrium. A hybrid straight-curved waveguide array is proposed to enable topological phase transitions from FCI to FATI and show how to diagnose the two phases using Bloch oscillations. As a proof of principle, the hybrid straight-curved waveguide array is designed as a straight honeycomb waveguide array nested in an asynchronous curved Kagome waveguide array. Under a two-dimensional (2D) tilted potential created by the spatial gradient of refractive indices, an initial Gaussian-like wavepacket undergoes 2D Bloch oscillations, displaying quasi-quantized displacement in the FCI and no drift in the FATI. This approach offers a direct and unambiguous method to diagnose Floquet topological phases from the bulk response.
2025-08-01
{'id': '2508.00582', 'title': 'Diagnosing Floquet Chern and anomalous topological insulators based on Bloch oscillations', 'abstract': 'It is challenging to distinguish Floquet Chern insulator (FCI) and Floquet\nanomalous topological insulator (FATI) because of their common features of\nchiral edge states and far away from equilibrium. A hybrid straight-curved\nwaveguide array is proposed to enable topological phase transitions from FCI to\nFATI and show how to diagnose the two phases using Bloch oscillations. As a\nproof of principle, the hybrid straight-curved waveguide array is designed as a\nstraight honeycomb waveguide array nested in an asynchronous curved Kagome\nwaveguide array. Under a two-dimensional (2D) tilted potential created by the\nspatial gradient of refractive indices, an initial Gaussian-like wavepacket\nundergoes 2D Bloch oscillations, displaying quasi-quantized displacement in the\nFCI and no drift in the FATI. This approach offers a direct and unambiguous\nmethod to diagnose Floquet topological phases from the bulk response.', 'authors': ['Maowu Zuo', 'Yongguan Ke', 'Zhoutao Lei', 'Chaohong Lee'], 'published': '2025-08-01 12:25:34+00:00', 'categories': ['physics.optics'], 'url': 'http://arxiv.org/abs/2508.00582v1'}
219
2508.00733
2,508.00733
null
AudioGen-Omni: A Unified Multimodal Diffusion Transformer for Video-Synchronized Audio, Speech, and Song Generation
We present AudioGen-Omni - a unified approach based on multimodal diffusion transformers (MMDit), capable of generating high-fidelity audio, speech, and song coherently synchronized with the input video. AudioGen-Omni introduces a novel joint training paradigm that seamlessly integrates large-scale video-text-audio corpora, enabling a model capable of generating semantically rich, acoustically diverse audio conditioned on multimodal inputs and adaptable to a wide range of audio generation tasks. AudioGen-Omni employs a unified lyrics-transcription encoder that encodes graphemes and phonemes from both song and spoken inputs into dense frame-level representations. Dense frame-level representations are fused using an AdaLN-based joint attention mechanism enhanced with phase-aligned anisotropic positional infusion (PAAPI), wherein RoPE is selectively applied to temporally structured modalities to ensure precise and robust cross-modal alignment. By unfreezing all modalities and masking missing inputs, AudioGen-Omni mitigates the semantic constraints of text-frozen paradigms, enabling effective cross-modal conditioning. This joint training approach enhances audio quality, semantic alignment, and lip-sync accuracy, while also achieving state-of-the-art results on Text-to-Audio/Speech/Song tasks. With an inference time of 1.91 seconds for 8 seconds of audio, it offers substantial improvements in both efficiency and generality.
2025-08-01
{'id': '2508.00733', 'title': 'AudioGen-Omni: A Unified Multimodal Diffusion Transformer for Video-Synchronized Audio, Speech, and Song Generation', 'abstract': 'We present AudioGen-Omni - a unified approach based on multimodal diffusion\ntransformers (MMDit), capable of generating high-fidelity audio, speech, and\nsong coherently synchronized with the input video. AudioGen-Omni introduces a\nnovel joint training paradigm that seamlessly integrates large-scale\nvideo-text-audio corpora, enabling a model capable of generating semantically\nrich, acoustically diverse audio conditioned on multimodal inputs and adaptable\nto a wide range of audio generation tasks. AudioGen-Omni employs a unified\nlyrics-transcription encoder that encodes graphemes and phonemes from both song\nand spoken inputs into dense frame-level representations. Dense frame-level\nrepresentations are fused using an AdaLN-based joint attention mechanism\nenhanced with phase-aligned anisotropic positional infusion (PAAPI), wherein\nRoPE is selectively applied to temporally structured modalities to ensure\nprecise and robust cross-modal alignment. By unfreezing all modalities and\nmasking missing inputs, AudioGen-Omni mitigates the semantic constraints of\ntext-frozen paradigms, enabling effective cross-modal conditioning. This joint\ntraining approach enhances audio quality, semantic alignment, and lip-sync\naccuracy, while also achieving state-of-the-art results on\nText-to-Audio/Speech/Song tasks. With an inference time of 1.91 seconds for 8\nseconds of audio, it offers substantial improvements in both efficiency and\ngenerality.', 'authors': ['Le Wang', 'Jun Wang', 'Chunyu Qiang', 'Feng Deng', 'Chen Zhang', 'Di Zhang', 'Kun Gai'], 'published': '2025-08-01 16:03:57+00:00', 'categories': ['cs.SD', 'cs.CV', 'cs.MM', 'eess.AS'], 'url': 'http://arxiv.org/abs/2508.00733v4'}
220
2508.00554
2,508.00554
null
ContestTrade: A Multi-Agent Trading System Based on Internal Contest Mechanism
In financial trading, large language model (LLM)-based agents demonstrate significant potential. However, the high sensitivity to market noise undermines the performance of LLM-based trading systems. To address this limitation, we propose a novel multi-agent system featuring an internal competitive mechanism inspired by modern corporate management structures. The system consists of two specialized teams: (1) Data Team - responsible for processing and condensing massive market data into diversified text factors, ensuring they fit the model's constrained context. (2) Research Team - tasked with making parallelized multipath trading decisions based on deep research methods. The core innovation lies in implementing a real-time evaluation and ranking mechanism within each team, driven by authentic market feedback. Each agent's performance undergoes continuous scoring and ranking, with only outputs from top-performing agents being adopted. The design enables the system to adaptively adjust to dynamic environment, enhances robustness against market noise and ultimately delivers superior trading performance. Experimental results demonstrate that our proposed system significantly outperforms prevailing multiagent systems and traditional quantitative investment methods across diverse evaluation metrics.
2025-08-01
{'id': '2508.00554', 'title': 'ContestTrade: A Multi-Agent Trading System Based on Internal Contest Mechanism', 'abstract': "In financial trading, large language model (LLM)-based agents demonstrate\nsignificant potential. However, the high sensitivity to market noise undermines\nthe performance of LLM-based trading systems. To address this limitation, we\npropose a novel multi-agent system featuring an internal competitive mechanism\ninspired by modern corporate management structures. The system consists of two\nspecialized teams: (1) Data Team - responsible for processing and condensing\nmassive market data into diversified text factors, ensuring they fit the\nmodel's constrained context. (2) Research Team - tasked with making\nparallelized multipath trading decisions based on deep research methods. The\ncore innovation lies in implementing a real-time evaluation and ranking\nmechanism within each team, driven by authentic market feedback. Each agent's\nperformance undergoes continuous scoring and ranking, with only outputs from\ntop-performing agents being adopted. The design enables the system to\nadaptively adjust to dynamic environment, enhances robustness against market\nnoise and ultimately delivers superior trading performance. Experimental\nresults demonstrate that our proposed system significantly outperforms\nprevailing multiagent systems and traditional quantitative investment methods\nacross diverse evaluation metrics.", 'authors': ['Li Zhao', 'Rui Sun', 'Zuoyou Jiang', 'Bo Yang', 'Yuxiao Bai', 'Mengting Chen', 'Xinyang Wang', 'Jing Li', 'Zuo Bai'], 'published': '2025-08-01 11:48:13+00:00', 'categories': ['q-fin.TR', 'cs.CL', 'q-fin.CP'], 'url': 'http://arxiv.org/abs/2508.00554v1'}
226
2508.00513
2,508.00513
null
Text-Attributed Graph Anomaly Detection via Multi-Scale Cross- and Uni-Modal Contrastive Learning
The widespread application of graph data in various high-risk scenarios has increased attention to graph anomaly detection (GAD). Faced with real-world graphs that often carry node descriptions in the form of raw text sequences, termed text-attributed graphs (TAGs), existing graph anomaly detection pipelines typically involve shallow embedding techniques to encode such textual information into features, and then rely on complex self-supervised tasks within the graph domain to detect anomalies. However, this text encoding process is separated from the anomaly detection training objective in the graph domain, making it difficult to ensure that the extracted textual features focus on GAD-relevant information, seriously constraining the detection capability. How to seamlessly integrate raw text and graph topology to unleash the vast potential of cross-modal data in TAGs for anomaly detection poses a challenging issue. This paper presents a novel end-to-end paradigm for text-attributed graph anomaly detection, named CMUCL. We simultaneously model data from both text and graph structures, and jointly train text and graph encoders by leveraging cross-modal and uni-modal multi-scale consistency to uncover potential anomaly-related information. Accordingly, we design an anomaly score estimator based on inconsistency mining to derive node-specific anomaly scores. Considering the lack of benchmark datasets tailored for anomaly detection on TAGs, we release 8 datasets to facilitate future research. Extensive evaluations show that CMUCL significantly advances in text-attributed graph anomaly detection, delivering an 11.13% increase in average accuracy (AP) over the suboptimal.
2025-08-01
{'id': '2508.00513', 'title': 'Text-Attributed Graph Anomaly Detection via Multi-Scale Cross- and Uni-Modal Contrastive Learning', 'abstract': 'The widespread application of graph data in various high-risk scenarios has\nincreased attention to graph anomaly detection (GAD). Faced with real-world\ngraphs that often carry node descriptions in the form of raw text sequences,\ntermed text-attributed graphs (TAGs), existing graph anomaly detection\npipelines typically involve shallow embedding techniques to encode such textual\ninformation into features, and then rely on complex self-supervised tasks\nwithin the graph domain to detect anomalies. However, this text encoding\nprocess is separated from the anomaly detection training objective in the graph\ndomain, making it difficult to ensure that the extracted textual features focus\non GAD-relevant information, seriously constraining the detection capability.\nHow to seamlessly integrate raw text and graph topology to unleash the vast\npotential of cross-modal data in TAGs for anomaly detection poses a challenging\nissue. This paper presents a novel end-to-end paradigm for text-attributed\ngraph anomaly detection, named CMUCL. We simultaneously model data from both\ntext and graph structures, and jointly train text and graph encoders by\nleveraging cross-modal and uni-modal multi-scale consistency to uncover\npotential anomaly-related information. Accordingly, we design an anomaly score\nestimator based on inconsistency mining to derive node-specific anomaly scores.\nConsidering the lack of benchmark datasets tailored for anomaly detection on\nTAGs, we release 8 datasets to facilitate future research. Extensive\nevaluations show that CMUCL significantly advances in text-attributed graph\nanomaly detection, delivering an 11.13% increase in average accuracy (AP) over\nthe suboptimal.', 'authors': ['Yiming Xu', 'Xu Hua', 'Zhen Peng', 'Bin Shi', 'Jiarun Chen', 'Xingbo Fu', 'Song Wang', 'Bo Dong'], 'published': '2025-08-01 10:47:07+00:00', 'categories': ['cs.LG'], 'url': 'http://arxiv.org/abs/2508.00513v1'}
221
2508.00612
2,508.00612
null
The substellar population in Corona Australis
The substellar initial mass function (IMF) and the formation mechanisms of brown dwarfs (BDs) remain key open questions in star formation theory. IMF characterization in a large number of star-forming regions (SFRs) is essential for constraining these processes. We aim to identify and spectroscopically confirm very low-mass members of the Corona Australis (CrA) SFR to refine its substellar census, determine its low-mass IMF, and compare it to other clusters. Using deep I-band photometry from SuprimeCam/Subaru and data from the VISTA Hemisphere Survey, we identified low-mass BD candidates in CrA. We subsequently obtained NIR spectra of 173 of these candidates with KMOS/VLT, as well as optical spectra for 8 kinematic candidate members using FLOYDS/LCO. The kinematic candidates are confirmed as low-mass stellar members with spectral types M1 to M5. In contrast, all 173 BD candidates observed with KMOS are identified as contaminants. Although the follow-up yielded no new substellar members, it places strong constraints on the number of undetected substellar objects in the region. Combined with literature data, this enables us to derive the substellar IMF, which is consistent with a single power-law slope of alpha = 0.95+-0.06 in the range 0.01-1 MSun or alpha = 0.33+-0.19 in the range 0.01-0.1 MSun. The star-to-BD ratio in CrA is about 2. We also provide updated IMFs and star-to-BD ratios for Lupus3 and ChaI from the SONYC survey, reflecting revised distances from Gaia. Finally, we estimate surface densities and median FUV fluxes for 6 SFRs and clusters and compare their substellar populations as a function of environmental properties. The IMF and star-to-BD ratio show no clear dependence on stellar density or ionizing flux from the massive stars. A combined effect - where one factor enhances and the other suppresses BD formation - also appears unlikely. (Abridged)
2025-08-01
{'id': '2508.00612', 'title': 'The substellar population in Corona Australis', 'abstract': 'The substellar initial mass function (IMF) and the formation mechanisms of\nbrown dwarfs (BDs) remain key open questions in star formation theory. IMF\ncharacterization in a large number of star-forming regions (SFRs) is essential\nfor constraining these processes. We aim to identify and spectroscopically\nconfirm very low-mass members of the Corona Australis (CrA) SFR to refine its\nsubstellar census, determine its low-mass IMF, and compare it to other\nclusters. Using deep I-band photometry from SuprimeCam/Subaru and data from the\nVISTA Hemisphere Survey, we identified low-mass BD candidates in CrA. We\nsubsequently obtained NIR spectra of 173 of these candidates with KMOS/VLT, as\nwell as optical spectra for 8 kinematic candidate members using FLOYDS/LCO. The\nkinematic candidates are confirmed as low-mass stellar members with spectral\ntypes M1 to M5. In contrast, all 173 BD candidates observed with KMOS are\nidentified as contaminants. Although the follow-up yielded no new substellar\nmembers, it places strong constraints on the number of undetected substellar\nobjects in the region. Combined with literature data, this enables us to derive\nthe substellar IMF, which is consistent with a single power-law slope of alpha\n= 0.95+-0.06 in the range 0.01-1 MSun or alpha = 0.33+-0.19 in the range\n0.01-0.1 MSun. The star-to-BD ratio in CrA is about 2. We also provide updated\nIMFs and star-to-BD ratios for Lupus3 and ChaI from the SONYC survey,\nreflecting revised distances from Gaia. Finally, we estimate surface densities\nand median FUV fluxes for 6 SFRs and clusters and compare their substellar\npopulations as a function of environmental properties. The IMF and star-to-BD\nratio show no clear dependence on stellar density or ionizing flux from the\nmassive stars. A combined effect - where one factor enhances and the other\nsuppresses BD formation - also appears unlikely. (Abridged)', 'authors': ['Koraljka Muzic', 'Víctor Almendros-Abad', 'André Baptista', 'Alexander Scholz', 'Daniel Capela', 'Samuel Pearson', 'Belinda Damian', 'Afonso doBrito-doVale', 'Tamara Rom', 'Ray Jayawardhana'], 'published': '2025-08-01 13:19:58+00:00', 'categories': ['astro-ph.SR', 'astro-ph.GA'], 'url': 'http://arxiv.org/abs/2508.00612v1'}
222
2508.00669
2,508.00669
null
Medical Reasoning in the Era of LLMs: A Systematic Review of Enhancement Techniques and Applications
The proliferation of Large Language Models (LLMs) in medicine has enabled impressive capabilities, yet a critical gap remains in their ability to perform systematic, transparent, and verifiable reasoning, a cornerstone of clinical practice. This has catalyzed a shift from single-step answer generation to the development of LLMs explicitly designed for medical reasoning. This paper provides the first systematic review of this emerging field. We propose a taxonomy of reasoning enhancement techniques, categorized into training-time strategies (e.g., supervised fine-tuning, reinforcement learning) and test-time mechanisms (e.g., prompt engineering, multi-agent systems). We analyze how these techniques are applied across different data modalities (text, image, code) and in key clinical applications such as diagnosis, education, and treatment planning. Furthermore, we survey the evolution of evaluation benchmarks from simple accuracy metrics to sophisticated assessments of reasoning quality and visual interpretability. Based on an analysis of 60 seminal studies from 2022-2025, we conclude by identifying critical challenges, including the faithfulness-plausibility gap and the need for native multimodal reasoning, and outlining future directions toward building efficient, robust, and sociotechnically responsible medical AI.
2025-08-01
{'id': '2508.00669', 'title': 'Medical Reasoning in the Era of LLMs: A Systematic Review of Enhancement Techniques and Applications', 'abstract': 'The proliferation of Large Language Models (LLMs) in medicine has enabled\nimpressive capabilities, yet a critical gap remains in their ability to perform\nsystematic, transparent, and verifiable reasoning, a cornerstone of clinical\npractice. This has catalyzed a shift from single-step answer generation to the\ndevelopment of LLMs explicitly designed for medical reasoning. This paper\nprovides the first systematic review of this emerging field. We propose a\ntaxonomy of reasoning enhancement techniques, categorized into training-time\nstrategies (e.g., supervised fine-tuning, reinforcement learning) and test-time\nmechanisms (e.g., prompt engineering, multi-agent systems). We analyze how\nthese techniques are applied across different data modalities (text, image,\ncode) and in key clinical applications such as diagnosis, education, and\ntreatment planning. Furthermore, we survey the evolution of evaluation\nbenchmarks from simple accuracy metrics to sophisticated assessments of\nreasoning quality and visual interpretability. Based on an analysis of 60\nseminal studies from 2022-2025, we conclude by identifying critical challenges,\nincluding the faithfulness-plausibility gap and the need for native multimodal\nreasoning, and outlining future directions toward building efficient, robust,\nand sociotechnically responsible medical AI.', 'authors': ['Wenxuan Wang', 'Zizhan Ma', 'Meidan Ding', 'Shiyi Zheng', 'Shengyuan Liu', 'Jie Liu', 'Jiaming Ji', 'Wenting Chen', 'Xiang Li', 'Linlin Shen', 'Yixuan Yuan'], 'published': '2025-08-01 14:41:31+00:00', 'categories': ['cs.CL', 'cs.AI', 'cs.CV', 'cs.LG'], 'url': 'http://arxiv.org/abs/2508.00669v1'}
223
2508.00231
2,508.00231
null
Generalizing the Penrose cut-and-paste method: Null shells with pressure and energy flux
The cut-and-paste method is a procedure for constructing null thin shells by matching two regions of the same spacetime across a null hypersurface. Originally proposed by Penrose, it has so far allowed to describe purely gravitational and null-dust shells in constant-curvature backgrounds. In this paper, we extend the cut-and-paste method to null shells with arbitrary gravitational/matter content. To that aim, we first derive a locally Lipschitz continuous form of the metric of the spacetime resulting from the most general matching of two constant-curvature spacetimes with totally geodesic null boundaries, and then obtain the coordinate transformation that turns this metric into the cut-and-paste form with a Dirac-delta term. The paper includes an example of a null shell with non-trivial energy density, energy flux and pressure in Minkowski space.
2025-08-01
{'id': '2508.00231', 'title': 'Generalizing the Penrose cut-and-paste method: Null shells with pressure and energy flux', 'abstract': 'The cut-and-paste method is a procedure for constructing null thin shells by\nmatching two regions of the same spacetime across a null hypersurface.\nOriginally proposed by Penrose, it has so far allowed to describe purely\ngravitational and null-dust shells in constant-curvature backgrounds. In this\npaper, we extend the cut-and-paste method to null shells with arbitrary\ngravitational/matter content. To that aim, we first derive a locally Lipschitz\ncontinuous form of the metric of the spacetime resulting from the most general\nmatching of two constant-curvature spacetimes with totally geodesic null\nboundaries, and then obtain the coordinate transformation that turns this\nmetric into the cut-and-paste form with a Dirac-delta term. The paper includes\nan example of a null shell with non-trivial energy density, energy flux and\npressure in Minkowski space.', 'authors': ['Miguel Manzano', 'Argam Ohanyan', 'Roland Steinbauer'], 'published': '2025-08-01 00:39:20+00:00', 'categories': ['math.DG', 'gr-qc', 'math-ph', 'math.MP', '53B30, 83C20, 83C25, 83C35'], 'url': 'http://arxiv.org/abs/2508.00231v1'}
224
2508.00573
2,508.00573
null
From Thermalization to Multifractality: Spin-Spin Correlation in Disordered $SU(2)$-Invariant 1D Heisenberg Spin Chains
We investigate spin correlations in one-dimensional $SU(2)$-invariant Heisenberg chains with exchange disorder for spin lengths $S=1/2$ and $S=1$. In the weak-disorder regime, the eigenmodes of the spin-spin correlation matrix are delocalized, consistent with ergodic behavior. Under strong disorder, the system enters a quasi-localized multifractal phase characterized by exponentially decaying, dimer-like spin correlations. Finite-size scaling of the inverse participation ratios of the correlation-matrix eigenmodes yields a correlation dimension, $D_20.37-0.39$, confirming the stability of a multifractal regime that is distinct from both the ergodic limit ($D_2=1$) and the fully localized limit ($D_2=0$).
2025-08-01
{'id': '2508.00573', 'title': 'From Thermalization to Multifractality: Spin-Spin Correlation in Disordered $SU(2)$-Invariant 1D Heisenberg Spin Chains', 'abstract': 'We investigate spin correlations in one-dimensional $SU(2)$-invariant\nHeisenberg chains with exchange disorder for spin lengths $S=1/2$ and $S=1$. In\nthe weak-disorder regime, the eigenmodes of the spin-spin correlation matrix\nare delocalized, consistent with ergodic behavior. Under strong disorder, the\nsystem enters a quasi-localized multifractal phase characterized by\nexponentially decaying, dimer-like spin correlations. Finite-size scaling of\nthe inverse participation ratios of the correlation-matrix eigenmodes yields a\ncorrelation dimension, $D_2\\approx 0.37-0.39$, confirming the stability of a\nmultifractal regime that is distinct from both the ergodic limit ($D_2=1$) and\nthe fully localized limit ($D_2=0$).', 'authors': ['Debasmita Giri', 'Julian Siegl', 'John Schliemann'], 'published': '2025-08-01 12:17:29+00:00', 'categories': ['cond-mat.dis-nn'], 'url': 'http://arxiv.org/abs/2508.00573v1'}
225
2508.00562
2,508.00562
null
Persistent Quantum Memory in Iterated Lifts
We study quantum coherence in continuous-time quantum walks on perfect graphs generated by the symmetric lift $HL'_2(G)$, a canonical, unweighted, undirected construction defined as the line graph of a bipartite double cover of $G$. This lift acts as both a coherence-preserving and coherence-inducing transformation: it preserves and scales structured quantum interference in highly symmetric base graphs, and induces sustained coherence in random or weakly structured ones. In small graphs such as $K_4$, $K_5$, and the Petersen graph, where quantum walks exhibit sharp revivals and high return probability, repeated $HL'_2$ lifting produces towers of perfect graphs with thousands to tens of thousands of vertices that retain periodic or quasi-periodic coherence. When applied to random regular or Erdos--R\'enyi graphs with flat or decaying return behavior, the lift introduces structured interference and significant amplification of mean and peak return probabilities. To quantify these effects, we evaluate standard coherence metrics from quantum resource theory, including inverse participation ratio (IPR), purity, relative entropy of coherence, and the logarithmic coherence number. These measures confirm that $HL'_2$ lifting delocalizes eigenstates, increases coherence entropy, and expands the basis support of quantum states. These results demonstrate that $HL'_2$ is a scalable and structurally grounded mechanism for organizing quantum interference, and introduce a new family of perfect graphs that support long-time quantum coherence without spectral tuning or engineered weights.
2025-08-01
{'id': '2508.00562', 'title': 'Persistent Quantum Memory in Iterated Lifts', 'abstract': "We study quantum coherence in continuous-time quantum walks on perfect graphs\ngenerated by the symmetric lift ${\\mathrm{HL}}'_2(G)$, a canonical, unweighted,\nundirected construction defined as the line graph of a bipartite double cover\nof $G$. This lift acts as both a coherence-preserving and coherence-inducing\ntransformation: it preserves and scales structured quantum interference in\nhighly symmetric base graphs, and induces sustained coherence in random or\nweakly structured ones.\n In small graphs such as $K_4$, $K_5$, and the Petersen graph, where quantum\nwalks exhibit sharp revivals and high return probability, repeated\n$\\mathrm{HL}'_2$ lifting produces towers of perfect graphs with thousands to\ntens of thousands of vertices that retain periodic or quasi-periodic coherence.\nWhen applied to random regular or Erd\\H{o}s--R\\'enyi graphs with flat or\ndecaying return behavior, the lift introduces structured interference and\nsignificant amplification of mean and peak return probabilities.\n To quantify these effects, we evaluate standard coherence metrics from\nquantum resource theory, including inverse participation ratio (IPR), purity,\nrelative entropy of coherence, and the logarithmic coherence number. These\nmeasures confirm that $\\mathrm{HL}'_2$ lifting delocalizes eigenstates,\nincreases coherence entropy, and expands the basis support of quantum states.\nThese results demonstrate that $\\mathrm{HL}'_2$ is a scalable and structurally\ngrounded mechanism for organizing quantum interference, and introduce a new\nfamily of perfect graphs that support long-time quantum coherence without\nspectral tuning or engineered weights.", 'authors': ['Hartosh Singh Bal'], 'published': '2025-08-01 12:02:05+00:00', 'categories': ['quant-ph', 'math.CO', 'es: 05C50 (Primary), 05C75, 05C90, 81P45, 81Q35 (Secondary)'], 'url': 'http://arxiv.org/abs/2508.00562v1'}
227
2508.00594
2,508.00594
null
Well-posedness of the periodic nonlinear Schrödinger equation with concentrated nonlinearity
We study the solution theory of the nonlinear Schr\"odinger equation with a concentrated nonlinearity on the torus. In particular, we establish existence and uniqueness of global energy-conserving solutions for small initial data in $H^1$. Our approach is based on two approximation schemes, namely the concentrated limit of a smoothed nonlinear Schr\"odinger equation and the inviscid limit of a concentrated complex Ginzburg--Landau equation. We also prove local well-posedness below the energy space. To our knowledge, this is the first rigorous solution theory for a periodic nonlinear Schr\"odinger equation with a concentrated nonlinearity.
2025-08-01
{'id': '2508.00594', 'title': 'Well-posedness of the periodic nonlinear Schrödinger equation with concentrated nonlinearity', 'abstract': 'We study the solution theory of the nonlinear Schr\\"odinger equation with a\nconcentrated nonlinearity on the torus. In particular, we establish existence\nand uniqueness of global energy-conserving solutions for small initial data in\n$H^1$. Our approach is based on two approximation schemes, namely the\nconcentrated limit of a smoothed nonlinear Schr\\"odinger equation and the\ninviscid limit of a concentrated complex Ginzburg--Landau equation. We also\nprove local well-posedness below the energy space. To our knowledge, this is\nthe first rigorous solution theory for a periodic nonlinear Schr\\"odinger\nequation with a concentrated nonlinearity.', 'authors': ['Jinyeop Lee', 'Andrew Rout'], 'published': '2025-08-01 12:49:51+00:00', 'categories': ['math.AP', 'math-ph', 'math.MP', '35Q55, 35B25, 35A01, 35A02, 45D05'], 'url': 'http://arxiv.org/abs/2508.00594v1'}
228
2508.00399
2,508.00399
null
iSafetyBench: A video-language benchmark for safety in industrial environment
Recent advances in vision-language models (VLMs) have enabled impressive generalization across diverse video understanding tasks under zero-shot settings. However, their capabilities in high-stakes industrial domains-where recognizing both routine operations and safety-critical anomalies is essential-remain largely underexplored. To address this gap, we introduce iSafetyBench, a new video-language benchmark specifically designed to evaluate model performance in industrial environments across both normal and hazardous scenarios. iSafetyBench comprises 1,100 video clips sourced from real-world industrial settings, annotated with open-vocabulary, multi-label action tags spanning 98 routine and 67 hazardous action categories. Each clip is paired with multiple-choice questions for both single-label and multi-label evaluation, enabling fine-grained assessment of VLMs in both standard and safety-critical contexts. We evaluate eight state-of-the-art video-language models under zero-shot conditions. Despite their strong performance on existing video benchmarks, these models struggle with iSafetyBench-particularly in recognizing hazardous activities and in multi-label scenarios. Our results reveal significant performance gaps, underscoring the need for more robust, safety-aware multimodal models for industrial applications. iSafetyBench provides a first-of-its-kind testbed to drive progress in this direction. The dataset is available at: https://github.com/raiyaan-abdullah/iSafety-Bench.
2025-08-01
{'id': '2508.00399', 'title': 'iSafetyBench: A video-language benchmark for safety in industrial environment', 'abstract': 'Recent advances in vision-language models (VLMs) have enabled impressive\ngeneralization across diverse video understanding tasks under zero-shot\nsettings. However, their capabilities in high-stakes industrial domains-where\nrecognizing both routine operations and safety-critical anomalies is\nessential-remain largely underexplored. To address this gap, we introduce\niSafetyBench, a new video-language benchmark specifically designed to evaluate\nmodel performance in industrial environments across both normal and hazardous\nscenarios. iSafetyBench comprises 1,100 video clips sourced from real-world\nindustrial settings, annotated with open-vocabulary, multi-label action tags\nspanning 98 routine and 67 hazardous action categories. Each clip is paired\nwith multiple-choice questions for both single-label and multi-label\nevaluation, enabling fine-grained assessment of VLMs in both standard and\nsafety-critical contexts. We evaluate eight state-of-the-art video-language\nmodels under zero-shot conditions. Despite their strong performance on existing\nvideo benchmarks, these models struggle with iSafetyBench-particularly in\nrecognizing hazardous activities and in multi-label scenarios. Our results\nreveal significant performance gaps, underscoring the need for more robust,\nsafety-aware multimodal models for industrial applications. iSafetyBench\nprovides a first-of-its-kind testbed to drive progress in this direction. The\ndataset is available at: https://github.com/raiyaan-abdullah/iSafety-Bench.', 'authors': ['Raiyaan Abdullah', 'Yogesh Singh Rawat', 'Shruti Vyas'], 'published': '2025-08-01 07:55:53+00:00', 'categories': ['cs.CV'], 'url': 'http://arxiv.org/abs/2508.00399v1'}
229
2508.00254
2,508.00254
null
Breakdown of Fermi's Golden Rule in 1d systems at non-zero temperature
In interacting quantum systems, the single-particle Green's function is expected to decay in time due to the interaction induced decoherence of quasiparticles. In the limit of weak interaction strengths ($$), a naive application of Fermi's Golden Rule (FGR) predicts an $O(^2)$ quasiparticle decay rate. However, for 1d fermions on the lattice at $T>0$, this calculation gives a divergent result and the scaling of the quasiparticle lifetime with interaction strength remains an open question. In this work we propose a solution to this question: combining numerical simulations using the recently introduced dissipation-assisted operator evolution (DAOE) method, with non-perturbative diagrammatic re-summations, we predict a logarithmic enhancement of the quasiparticle decay rate $^-1 ^2 ^-2$. We argue that this effect is present in a wide variety of well-known weakly interacting quantum fermionic and bosonic systems, and even in some classical systems, provided the non-interacting limit has quasiparticles with a generic dispersion.
2025-08-01
{'id': '2508.00254', 'title': "Breakdown of Fermi's Golden Rule in 1d systems at non-zero temperature", 'abstract': "In interacting quantum systems, the single-particle Green's function is\nexpected to decay in time due to the interaction induced decoherence of\nquasiparticles. In the limit of weak interaction strengths ($\\Delta$), a naive\napplication of Fermi's Golden Rule (FGR) predicts an $\\mathcal{O}(\\Delta^{2})$\nquasiparticle decay rate. However, for 1d fermions on the lattice at $T>0$,\nthis calculation gives a divergent result and the scaling of the quasiparticle\nlifetime with interaction strength remains an open question. In this work we\npropose a solution to this question: combining numerical simulations using the\nrecently introduced dissipation-assisted operator evolution (DAOE) method, with\nnon-perturbative diagrammatic re-summations, we predict a logarithmic\nenhancement of the quasiparticle decay rate $\\tau^{-1} \\sim \\Delta^{2} \\log\n\\Delta^{-2}$. We argue that this effect is present in a wide variety of\nwell-known weakly interacting quantum fermionic and bosonic systems, and even\nin some classical systems, provided the non-interacting limit has\nquasiparticles with a generic dispersion.", 'authors': ['Thomas Young', 'Jerome LLoyd', 'Curt von Keyserlingk'], 'published': '2025-08-01 01:51:50+00:00', 'categories': ['quant-ph', 'cond-mat.quant-gas', 'cond-mat.str-el'], 'url': 'http://arxiv.org/abs/2508.00254v1'}
230
2508.00245
2,508.00245
null
The Subaru-Asahi StarCam: Description of the system
The Subaru-Asahi StarCam is a high-sensitivity live-streaming camera for meteor observation, installed on the dome of the Subaru Telescope at the summit area of Maunakea, Hawai'i. Although it was originally intended to share the Maunakea night sky with the public, including the local Hawai'i community, the system quickly demonstrated its potential for scientific research, owing to its highly sensitive video capabilities and the exceptional fraction of clear nights at the site. The core of the StarCam system features a Sony FX3 camera body paired with an F1.4 wide-angle lens, offering a field of view of 70 deg by 40 deg. Leveraging a state-of-the-art, high-sensitivity CMOS sensor and a bright lens, the system is capable of capturing stars as faint as magnitude 8 in real-time, with an effective frame rate of 15--30 fps. Live streaming via YouTube began in April 2021, and the feed is constantly monitored by more than a hundred viewers at any given nighttime. This has enabled the camera to be used not only for observing regular meteor showers but also for monitoring scientifically important phenomena such as fireballs or unexpected meteor outbursts. Notable scientific achievements include: 1) Detection of the new Arid meteor shower in 2021, 2) Identification of a sub-peak activity in the Gamma-Perseid meteor shower (2021), 3) Detection of the 2022 Tau-Herculid meteor shower outburst, 4) Confirmation of the activity of the Andromedid meteor shower (2021), and 5) Multiple detections of meteor cluster phenomena. We discuss the potential and the future scope of StarCam as an open-access, real-time data platform for citizen science in meteor observations.
2025-08-01
{'id': '2508.00245', 'title': 'The Subaru-Asahi StarCam: Description of the system', 'abstract': "The Subaru-Asahi StarCam is a high-sensitivity live-streaming camera for\nmeteor observation, installed on the dome of the Subaru Telescope at the summit\narea of Maunakea, Hawai'i. Although it was originally intended to share the\nMaunakea night sky with the public, including the local Hawai'i community, the\nsystem quickly demonstrated its potential for scientific research, owing to its\nhighly sensitive video capabilities and the exceptional fraction of clear\nnights at the site. The core of the StarCam system features a Sony FX3 camera\nbody paired with an F1.4 wide-angle lens, offering a field of view of 70 deg by\n40 deg. Leveraging a state-of-the-art, high-sensitivity CMOS sensor and a\nbright lens, the system is capable of capturing stars as faint as magnitude 8\nin real-time, with an effective frame rate of 15--30 fps. Live streaming via\nYouTube began in April 2021, and the feed is constantly monitored by more than\na hundred viewers at any given nighttime. This has enabled the camera to be\nused not only for observing regular meteor showers but also for monitoring\nscientifically important phenomena such as fireballs or unexpected meteor\noutbursts. Notable scientific achievements include: 1) Detection of the new\nArid meteor shower in 2021, 2) Identification of a sub-peak activity in the\nGamma-Perseid meteor shower (2021), 3) Detection of the 2022 Tau-Herculid\nmeteor shower outburst, 4) Confirmation of the activity of the Andromedid\nmeteor shower (2021), and 5) Multiple detections of meteor cluster phenomena.\nWe discuss the potential and the future scope of StarCam as an open-access,\nreal-time data platform for citizen science in meteor observations.", 'authors': ['Ichi Tanaka', 'Masanobu Higashiyama', 'Masayo Nakajima', 'Toyokazu Uda', 'Hitoshi Hasegawa', 'Mikiya Sato', 'Jun-ichi Watanabe'], 'published': '2025-08-01 01:06:44+00:00', 'categories': ['astro-ph.IM', 'astro-ph.EP'], 'url': 'http://arxiv.org/abs/2508.00245v1'}
236
2508.00645
2,508.00645
null
SmartFlow: A CFD-solver-agnostic deep reinforcement learning framework for computational fluid dynamics on HPC platforms
Deep reinforcement learning (DRL) is emerging as a powerful tool for fluid-dynamics research, encompassing active flow control, autonomous navigation, turbulence modeling and discovery of novel numerical schemes. We introduce SmartFlow, a CFD-solver-agnostic framework for both single- and multi-agent DRL algorithms that can easily integrate with MPI-parallel CPU and GPU-accelerated solvers. Built on Relexi and SmartSOD2D, SmartFlow uses the SmartSim infrastructure library and our newly developed SmartRedis-MPI library to enable asynchronous, low-latency, in-memory communication between CFD solvers and Python-based DRL algorithms. SmartFlow leverages PyTorch's Stable-Baselines3 for training, which provides a modular, Gym-like environment API. We demonstrate its versatility via three case studies: single-agent synthetic-jet control for drag reduction in a cylinder flow simulated by the high-order FLEXI solver, multi-agent cylinder wake control using the GPU-accelerated spectral-element code SOD2D, and multi-agent wall-model learning for large-eddy simulation with the finite-difference solver CaLES. SmartFlow's CFD-solver-agnostic design and seamless HPC integration is promising to accelerate RL-driven fluid-mechanics studies.
2025-08-01
{'id': '2508.00645', 'title': 'SmartFlow: A CFD-solver-agnostic deep reinforcement learning framework for computational fluid dynamics on HPC platforms', 'abstract': "Deep reinforcement learning (DRL) is emerging as a powerful tool for\nfluid-dynamics research, encompassing active flow control, autonomous\nnavigation, turbulence modeling and discovery of novel numerical schemes. We\nintroduce SmartFlow, a CFD-solver-agnostic framework for both single- and\nmulti-agent DRL algorithms that can easily integrate with MPI-parallel CPU and\nGPU-accelerated solvers. Built on Relexi and SmartSOD2D, SmartFlow uses the\nSmartSim infrastructure library and our newly developed SmartRedis-MPI library\nto enable asynchronous, low-latency, in-memory communication between CFD\nsolvers and Python-based DRL algorithms. SmartFlow leverages PyTorch's\nStable-Baselines3 for training, which provides a modular, Gym-like environment\nAPI. We demonstrate its versatility via three case studies: single-agent\nsynthetic-jet control for drag reduction in a cylinder flow simulated by the\nhigh-order FLEXI solver, multi-agent cylinder wake control using the\nGPU-accelerated spectral-element code SOD2D, and multi-agent wall-model\nlearning for large-eddy simulation with the finite-difference solver CaLES.\nSmartFlow's CFD-solver-agnostic design and seamless HPC integration is\npromising to accelerate RL-driven fluid-mechanics studies.", 'authors': ['Maochao Xiao', 'Yuning Wang', 'Felix Rodach', 'Bernat Font', 'Marius Kurz', 'Pol Suárez', 'Di Zhou', 'Francisco Alcántara-Ávila', 'Ting Zhu', 'Junle Liu', 'Ricard Montalà', 'Jiawei Chen', 'Jean Rabault', 'Oriol Lehmkuhl', 'Andrea Beck', 'Johan Larsson', 'Ricardo Vinuesa', 'Sergio Pirozzoli'], 'published': '2025-08-01 14:03:41+00:00', 'categories': ['physics.flu-dyn', 'physics.comp-ph'], 'url': 'http://arxiv.org/abs/2508.00645v1'}
231
2508.00578
2,508.00578
null
Learning Potential Energy Surfaces of Hydrogen Atom Transfer Reactions in Peptides
Hydrogen atom transfer (HAT) reactions are essential in many biological processes, such as radical migration in damaged proteins, but their mechanistic pathways remain incompletely understood. Simulating HAT is challenging due to the need for quantum chemical accuracy at biologically relevant scales; thus, neither classical force fields nor DFT-based molecular dynamics are applicable. Machine-learned potentials offer an alternative, able to learn potential energy surfaces (PESs) with near-quantum accuracy. However, training these models to generalize across diverse HAT configurations, especially at radical positions in proteins, requires tailored data generation and careful model selection. Here, we systematically generate HAT configurations in peptides to build large datasets using semiempirical methods and DFT. We benchmark three graph neural network architectures (SchNet, Allegro, and MACE) on their ability to learn HAT PESs and indirectly predict reaction barriers from energy predictions. MACE consistently outperforms the others in energy, force, and barrier prediction, achieving a mean absolute error of 1.13 kcal/mol on out-of-distribution DFT barrier predictions. This accuracy enables integration of ML potentials into large-scale collagen simulations to compute reaction rates from predicted barriers, advancing mechanistic understanding of HAT and radical migration in peptides. We analyze scaling laws, model transferability, and cost-performance trade-offs, and outline strategies for improvement by combining ML potentials with transition state search algorithms and active learning. Our approach is generalizable to other biomolecular systems, enabling quantum-accurate simulations of chemical reactivity in complex environments.
2025-08-01
{'id': '2508.00578', 'title': 'Learning Potential Energy Surfaces of Hydrogen Atom Transfer Reactions in Peptides', 'abstract': 'Hydrogen atom transfer (HAT) reactions are essential in many biological\nprocesses, such as radical migration in damaged proteins, but their mechanistic\npathways remain incompletely understood. Simulating HAT is challenging due to\nthe need for quantum chemical accuracy at biologically relevant scales; thus,\nneither classical force fields nor DFT-based molecular dynamics are applicable.\nMachine-learned potentials offer an alternative, able to learn potential energy\nsurfaces (PESs) with near-quantum accuracy. However, training these models to\ngeneralize across diverse HAT configurations, especially at radical positions\nin proteins, requires tailored data generation and careful model selection.\nHere, we systematically generate HAT configurations in peptides to build large\ndatasets using semiempirical methods and DFT. We benchmark three graph neural\nnetwork architectures (SchNet, Allegro, and MACE) on their ability to learn HAT\nPESs and indirectly predict reaction barriers from energy predictions. MACE\nconsistently outperforms the others in energy, force, and barrier prediction,\nachieving a mean absolute error of 1.13 kcal/mol on out-of-distribution DFT\nbarrier predictions. This accuracy enables integration of ML potentials into\nlarge-scale collagen simulations to compute reaction rates from predicted\nbarriers, advancing mechanistic understanding of HAT and radical migration in\npeptides. We analyze scaling laws, model transferability, and cost-performance\ntrade-offs, and outline strategies for improvement by combining ML potentials\nwith transition state search algorithms and active learning. Our approach is\ngeneralizable to other biomolecular systems, enabling quantum-accurate\nsimulations of chemical reactivity in complex environments.', 'authors': ['Marlen Neubert', 'Patrick Reiser', 'Frauke Gräter', 'Pascal Friederich'], 'published': '2025-08-01 12:21:49+00:00', 'categories': ['cs.LG', 'cond-mat.mtrl-sci', 'physics.chem-ph', 'physics.comp-ph', 'q-bio.BM'], 'url': 'http://arxiv.org/abs/2508.00578v1'}
232
2508.00302
2,508.00302
null
The net-regular strongly regular signed graphs with degree 6
In this paper, we study the net-regular strongly regular signed graphs with degree 6 and determine all connected 6-regular and net-regular strongly regular signed graphs. There are three, six and four 6-regular strongly regular signed graphs with net-degree 4, 2 and 0, respectively.
2025-08-01
{'id': '2508.00302', 'title': 'The net-regular strongly regular signed graphs with degree 6', 'abstract': 'In this paper, we study the net-regular strongly regular signed graphs with\ndegree 6 and determine all connected 6-regular and net-regular strongly regular\nsigned graphs. There are three, six and four 6-regular strongly regular signed\ngraphs with net-degree 4, 2 and 0, respectively.', 'authors': ['Qian Yu', 'Yaoping Hou'], 'published': '2025-08-01 04:07:57+00:00', 'categories': ['math.CO'], 'url': 'http://arxiv.org/abs/2508.00302v1'}
233
2508.00672
2,508.00672
null
On type 1 active galactic nuclei with double-peaked [O~{\sc iii}]. II. properties of long-term optical variability
Double-peaked ~profiles could potentially indicate kiloparsec-scale dual AGNs. We analyze long-term optical light curves of 35 type 1 AGNs with such features from our recent catalog in Zheng et al. (2025). These light curves are obtained from the Catalina Sky Survey and modeled using a Damped Random Walk (DRW) process. A control sample of 210 normal type 1 AGNs matched in redshift, intrinsic luminosity, and black hole mass is also studied. If the double-peaked ~are caused by two type 1 AGNs (dual type 1 AGN), then the combined variability from the two AGNs would be expected to differ from that of a single type 1 AGN. However, there is no statistically significant difference in the variability timescale $$ and intrinsic variability amplitude $$ between these double-peaked AGNs and the control sample of 210 normal type 1 AGNs. Crucially, computer simulations reveal that dual AGN systems systematically produce lower variability amplitudes than single AGNs, which is inconsistent with the observed variability properties of our double-peaked ~sample. Moreover, simulations suggest that the fraction of dual type 1 AGNs is $$ 3\%, indicating that double-peaked ~may not be a reliable indicator of dual type 1 AGNs in these systems. However, this does not rule out the possibility that some objects may still host dual AGNs involving other combinations, such as type 1+type 2 AGNs. Future studies with larger samples and higher-quality light curves will help clarify the true nature of these systems.
2025-08-01
{'id': '2508.00672', 'title': 'On type 1 active galactic nuclei with double-peaked [O~{\\sc iii}]. II. properties of long-term optical variability', 'abstract': 'Double-peaked \\oiii~profiles could potentially indicate kiloparsec-scale dual\nAGNs. We analyze long-term optical light curves of 35 type 1 AGNs with such\nfeatures from our recent catalog in Zheng et al. (2025). These light curves are\nobtained from the Catalina Sky Survey and modeled using a Damped Random Walk\n(DRW) process. A control sample of 210 normal type 1 AGNs matched in redshift,\nintrinsic luminosity, and black hole mass is also studied. If the double-peaked\n\\oiii~are caused by two type 1 AGNs (dual type 1 AGN), then the combined\nvariability from the two AGNs would be expected to differ from that of a single\ntype 1 AGN. However, there is no statistically significant difference in the\nvariability timescale $\\tau$ and intrinsic variability amplitude $\\sigma$\nbetween these double-peaked AGNs and the control sample of 210 normal type 1\nAGNs. Crucially, computer simulations reveal that dual AGN systems\nsystematically produce lower variability amplitudes than single AGNs, which is\ninconsistent with the observed variability properties of our double-peaked\n\\oiii~sample. Moreover, simulations suggest that the fraction of dual type 1\nAGNs is $\\sim$ 3\\%, indicating that double-peaked \\oiii~may not be a reliable\nindicator of dual type 1 AGNs in these systems. However, this does not rule out\nthe possibility that some objects may still host dual AGNs involving other\ncombinations, such as type 1+type 2 AGNs. Future studies with larger samples\nand higher-quality light curves will help clarify the true nature of these\nsystems.', 'authors': ['Qi Zheng', 'Xingyv Zhu', 'Xueguang Zhang', 'Qirong Yuan'], 'published': '2025-08-01 14:46:47+00:00', 'categories': ['astro-ph.GA'], 'url': 'http://arxiv.org/abs/2508.00672v1'}
234
2508.00546
2,508.00546
null
SPENCER: Self-Adaptive Model Distillation for Efficient Code Retrieval
Code retrieval aims to provide users with desired code snippets based on users' natural language queries. With the development of deep learning technologies, adopting pre-trained models for this task has become mainstream. Considering the retrieval efficiency, most of the previous approaches adopt a dual-encoder for this task, which encodes the description and code snippet into representation vectors, respectively. However, the model structure of the dual-encoder tends to limit the model's performance, since it lacks the interaction between the code snippet and description at the bottom layer of the model during training. To improve the model's effectiveness while preserving its efficiency, we propose a framework, which adopts Self-AdaPtive Model Distillation for Efficient CodE Retrieval, named SPENCER. SPENCER first adopts the dual-encoder to narrow the search space and then adopts the cross-encoder to improve accuracy. To improve the efficiency of SPENCER, we propose a novel model distillation technique, which can greatly reduce the inference time of the dual-encoder while maintaining the overall performance. We also propose a teaching assistant selection strategy for our model distillation, which can adaptively select the suitable teaching assistant models for different pre-trained models during the model distillation to ensure the model performance. Extensive experiments demonstrate that the combination of dual-encoder and cross-encoder improves overall performance compared to solely dual-encoder-based models for code retrieval. Besides, our model distillation technique retains over 98% of the overall performance while reducing the inference time of the dual-encoder by 70%.
2025-08-01
{'id': '2508.00546', 'title': 'SPENCER: Self-Adaptive Model Distillation for Efficient Code Retrieval', 'abstract': "Code retrieval aims to provide users with desired code snippets based on\nusers' natural language queries. With the development of deep learning\ntechnologies, adopting pre-trained models for this task has become mainstream.\nConsidering the retrieval efficiency, most of the previous approaches adopt a\ndual-encoder for this task, which encodes the description and code snippet into\nrepresentation vectors, respectively. However, the model structure of the\ndual-encoder tends to limit the model's performance, since it lacks the\ninteraction between the code snippet and description at the bottom layer of the\nmodel during training. To improve the model's effectiveness while preserving\nits efficiency, we propose a framework, which adopts Self-AdaPtive Model\nDistillation for Efficient CodE Retrieval, named SPENCER. SPENCER first adopts\nthe dual-encoder to narrow the search space and then adopts the cross-encoder\nto improve accuracy. To improve the efficiency of SPENCER, we propose a novel\nmodel distillation technique, which can greatly reduce the inference time of\nthe dual-encoder while maintaining the overall performance. We also propose a\nteaching assistant selection strategy for our model distillation, which can\nadaptively select the suitable teaching assistant models for different\npre-trained models during the model distillation to ensure the model\nperformance. Extensive experiments demonstrate that the combination of\ndual-encoder and cross-encoder improves overall performance compared to solely\ndual-encoder-based models for code retrieval. Besides, our model distillation\ntechnique retains over 98% of the overall performance while reducing the\ninference time of the dual-encoder by 70%.", 'authors': ['Wenchao Gu', 'Zongyi Lyu', 'Yanlin Wang', 'Hongyu Zhang', 'Cuiyun Gao', 'Michael R. Lyu'], 'published': '2025-08-01 11:39:32+00:00', 'categories': ['cs.SE', 'cs.AI'], 'url': 'http://arxiv.org/abs/2508.00546v1'}
235
2508.00564
2,508.00564
null
Aspects of 4d $\mathcal{N}=1$ $ADE$ gauge theories from M-theory: decomposition, automorphisms, and generalised symmetries
We study the decomposition of 4d $N=1$ gauge theories with Lie algebras of type $su(N)$, $so(2N)$, and $e_6$, realized via M-theory geometric engineering. These theories, together with their novel decomposition structure, arise from quotienting the Bryant--Salamon spin bundle over the 3-sphere by special finite subgroups acting simultaneously on both the fiber and base. We show that these gauge theories admit both inner and outer automorphisms, enabling sequences of gauge theory breaking. In particular, outer automorphisms extend the decomposition structure to theories with $so(2N+1)$, $sp(2N)$, $f_4$, and $g_2$ gauge algebras. For these theories, including both simply-laced and non-simply-laced cases, we analyze their $p$-form symmetries, including $(-1)$-form symmetries, derive the corresponding SymTFTs, and identify the M-theoretic origin of their symmetry topological operators and defects. Finally, we demonstrate that these gauge theories exhibit modified instanton sums and higher 4-group structures, and we derive the associated topological sector directly from M-theory.
2025-08-01
{'id': '2508.00564', 'title': 'Aspects of 4d $\\mathcal{N}=1$ $ADE$ gauge theories from M-theory: decomposition, automorphisms, and generalised symmetries', 'abstract': 'We study the decomposition of 4d $\\mathcal{N}=1$ gauge theories with Lie\nalgebras of type $\\mathfrak{su}(N)$, $\\mathfrak{so}(2N)$, and\n$\\mathfrak{e}_{6}$, realized via M-theory geometric engineering. These\ntheories, together with their novel decomposition structure, arise from\nquotienting the Bryant--Salamon spin bundle over the 3-sphere by special finite\nsubgroups acting simultaneously on both the fiber and base. We show that these\ngauge theories admit both inner and outer automorphisms, enabling sequences of\ngauge theory breaking. In particular, outer automorphisms extend the\ndecomposition structure to theories with $\\mathfrak{so}(2N+1)$,\n$\\mathfrak{sp}(2N)$, $\\mathfrak{f}_{4}$, and $\\mathfrak{g}_{2}$ gauge algebras.\nFor these theories, including both simply-laced and non-simply-laced cases, we\nanalyze their $p$-form symmetries, including $(-1)$-form symmetries, derive the\ncorresponding SymTFTs, and identify the M-theoretic origin of their symmetry\ntopological operators and defects. Finally, we demonstrate that these gauge\ntheories exhibit modified instanton sums and higher 4-group structures, and we\nderive the associated topological sector directly from M-theory.', 'authors': ['Osama Khlaif', 'Marwan Najjar'], 'published': '2025-08-01 12:04:49+00:00', 'categories': ['hep-th'], 'url': 'http://arxiv.org/abs/2508.00564v1'}
237
2508.00635
2,508.00635
null
KFS: KAN based adaptive Frequency Selection learning architecture for long term time series forecasting
Multi-scale decomposition architectures have emerged as predominant methodologies in time series forecasting. However, real-world time series exhibit noise interference across different scales, while heterogeneous information distribution among frequency components at varying scales leads to suboptimal multi-scale representation. Inspired by Kolmogorov-Arnold Networks (KAN) and Parseval's theorem, we propose a KAN based adaptive Frequency Selection learning architecture (KFS) to address these challenges. This framework tackles prediction challenges stemming from cross-scale noise interference and complex pattern modeling through its FreK module, which performs energy-distribution-based dominant frequency selection in the spectral domain. Simultaneously, KAN enables sophisticated pattern representation while timestamp embedding alignment synchronizes temporal representations across scales. The feature mixing module then fuses scale-specific patterns with aligned temporal features. Extensive experiments across multiple real-world time series datasets demonstrate that KT achieves state-of-the-art performance as a simple yet effective architecture.
2025-08-01
{'id': '2508.00635', 'title': 'KFS: KAN based adaptive Frequency Selection learning architecture for long term time series forecasting', 'abstract': "Multi-scale decomposition architectures have emerged as predominant\nmethodologies in time series forecasting. However, real-world time series\nexhibit noise interference across different scales, while heterogeneous\ninformation distribution among frequency components at varying scales leads to\nsuboptimal multi-scale representation. Inspired by Kolmogorov-Arnold Networks\n(KAN) and Parseval's theorem, we propose a KAN based adaptive Frequency\nSelection learning architecture (KFS) to address these challenges. This\nframework tackles prediction challenges stemming from cross-scale noise\ninterference and complex pattern modeling through its FreK module, which\nperforms energy-distribution-based dominant frequency selection in the spectral\ndomain. Simultaneously, KAN enables sophisticated pattern representation while\ntimestamp embedding alignment synchronizes temporal representations across\nscales. The feature mixing module then fuses scale-specific patterns with\naligned temporal features. Extensive experiments across multiple real-world\ntime series datasets demonstrate that KT achieves state-of-the-art performance\nas a simple yet effective architecture.", 'authors': ['Changning Wu', 'Gao Wu', 'Rongyao Cai', 'Yong Liu', 'Kexin Zhang'], 'published': '2025-08-01 13:50:57+00:00', 'categories': ['cs.LG'], 'url': 'http://arxiv.org/abs/2508.00635v2'}
238
2508.00595
2,508.00595
null
The standard construction for cocycle twisted and braided tensor product W$^*$-algebras
Given a locally compact quantum group $G$ and a (generalized) dual unitary $2$-cocycle $$, any W$^*$-algebra $A$ with a $G$-action can be twisted into a new W$^*$-algebra $A_$ with an action by the cocycle twist $G_$ of $G$. We show how, in general, the standard space $L^2(A_)$, with its standard $G_$-representation, can be seen as a twist of $L^2(A)$ with its standard $G$-representation. We then apply this general result in the special case of (generalized) Drinfeld doubles.
2025-08-01
{'id': '2508.00595', 'title': 'The standard construction for cocycle twisted and braided tensor product W$^*$-algebras', 'abstract': 'Given a locally compact quantum group $\\mathbb{G}$ and a (generalized) dual\nunitary $2$-cocycle $\\hat{\\Omega}$, any W$^*$-algebra $A$ with a\n$\\mathbb{G}$-action can be twisted into a new W$^*$-algebra $A_{\\hat{\\Omega}}$\nwith an action by the cocycle twist $\\mathbb{G}_{\\hat{\\Omega}}$ of\n$\\mathbb{G}$. We show how, in general, the standard space\n$L^2(A_{\\hat{\\Omega}})$, with its standard\n$\\mathbb{G}_{\\hat{\\Omega}}$-representation, can be seen as a twist of $L^2(A)$\nwith its standard $\\mathbb{G}$-representation. We then apply this general\nresult in the special case of (generalized) Drinfeld doubles.', 'authors': ['K. De Commer', 'J. Krajczok'], 'published': '2025-08-01 12:50:29+00:00', 'categories': ['math.OA', 'Primary 46L67, Secondary 46L55, 46L06'], 'url': 'http://arxiv.org/abs/2508.00595v1'}
239
2508.00430
2,508.0043
null
The effect of dephasing and spin-lattice relaxation during the switching processes in quantum antiferromagnets
The control of antiferromagnetic order can pave the way to large storage capacity as well as fast manipulation of stored data. Here achieving a steady-state of sublattice magnetization after switching is crucial to prevent loss of stored data. The present theoretical approach aims to obtain instantaneous stable states of the order after reorienting the N\'eel vector in open quantum antiferromagnets using time-dependent Schwinger boson mean-field theory. The Lindblad formalism is employed to couple the system to the environment. The quantum theoretical approach comprises differences in the effects of dephasing, originating from destructive interference of different wave vectors, and spin-lattice relaxation. We show that the spin-lattice relaxation results in an exponentially fast convergence to the steady-state after full ultrafast switching.
2025-08-01
{'id': '2508.00430', 'title': 'The effect of dephasing and spin-lattice relaxation during the switching processes in quantum antiferromagnets', 'abstract': "The control of antiferromagnetic order can pave the way to large storage\ncapacity as well as fast manipulation of stored data. Here achieving a\nsteady-state of sublattice magnetization after switching is crucial to prevent\nloss of stored data. The present theoretical approach aims to obtain\ninstantaneous stable states of the order after reorienting the N\\'eel vector in\nopen quantum antiferromagnets using time-dependent Schwinger boson mean-field\ntheory. The Lindblad formalism is employed to couple the system to the\nenvironment. The quantum theoretical approach comprises differences in the\neffects of dephasing, originating from destructive interference of different\nwave vectors, and spin-lattice relaxation. We show that the spin-lattice\nrelaxation results in an exponentially fast convergence to the steady-state\nafter full ultrafast switching.", 'authors': ['Asliddin Khudoyberdiev', 'Götz S. Uhrig'], 'published': '2025-08-01 08:39:27+00:00', 'categories': ['cond-mat.mtrl-sci', 'cond-mat.str-el'], 'url': 'http://arxiv.org/abs/2508.00430v1'}
240
2508.00429
2,508.00429
null
ReaGAN: Node-as-Agent-Reasoning Graph Agentic Network
Graph Neural Networks (GNNs) have achieved remarkable success in graph-based learning by propagating information among neighbor nodes via predefined aggregation mechanisms. However, such fixed schemes often suffer from two key limitations. First, they cannot handle the imbalance in node informativeness -- some nodes are rich in information, while others remain sparse. Second, predefined message passing primarily leverages local structural similarity while ignoring global semantic relationships across the graph, limiting the model's ability to capture distant but relevant information. We propose Retrieval-augmented Graph Agentic Network (ReaGAN), an agent-based framework that empowers each node with autonomous, node-level decision-making. Each node acts as an agent that independently plans its next action based on its internal memory, enabling node-level planning and adaptive message propagation. Additionally, retrieval-augmented generation (RAG) allows nodes to access semantically relevant content and build global relationships in the graph. ReaGAN achieves competitive performance under few-shot in-context settings using a frozen LLM backbone without fine-tuning, showcasing the potential of agentic planning and local-global retrieval in graph learning.
2025-08-01
{'id': '2508.00429', 'title': 'ReaGAN: Node-as-Agent-Reasoning Graph Agentic Network', 'abstract': "Graph Neural Networks (GNNs) have achieved remarkable success in graph-based\nlearning by propagating information among neighbor nodes via predefined\naggregation mechanisms. However, such fixed schemes often suffer from two key\nlimitations. First, they cannot handle the imbalance in node informativeness --\nsome nodes are rich in information, while others remain sparse. Second,\npredefined message passing primarily leverages local structural similarity\nwhile ignoring global semantic relationships across the graph, limiting the\nmodel's ability to capture distant but relevant information. We propose\nRetrieval-augmented Graph Agentic Network (ReaGAN), an agent-based framework\nthat empowers each node with autonomous, node-level decision-making. Each node\nacts as an agent that independently plans its next action based on its internal\nmemory, enabling node-level planning and adaptive message propagation.\nAdditionally, retrieval-augmented generation (RAG) allows nodes to access\nsemantically relevant content and build global relationships in the graph.\nReaGAN achieves competitive performance under few-shot in-context settings\nusing a frozen LLM backbone without fine-tuning, showcasing the potential of\nagentic planning and local-global retrieval in graph learning.", 'authors': ['Minghao Guo', 'Xi Zhu', 'Jingyuan Huang', 'Kai Mei', 'Yongfeng Zhang'], 'published': '2025-08-01 08:37:54+00:00', 'categories': ['cs.CL', 'cs.LG', 'cs.MA'], 'url': 'http://arxiv.org/abs/2508.00429v2'}
241
2508.00597
2,508.00597
null
Biproduct Quasi-Hopf Algebras of Rank 2
Inspired by the work of Radford, for $H$ an arbitrary quasi-Hopf algebra we describe all the Hopf algebras of dimension $2$ within the braided category of left Yetter-Drinfeld modules over $H$ and determine the biproduct quasi-Hopf algebras defined by them. Classes of such biproduct quasi-Hopf algebras are obtained by taking $H$ as the Hopf algebra of functions on a group $G$, endowed with the quasi-Hopf algebra structure provided by a non-trivial $3$-cocycle on $G$ (especially when $G$ is a finite cyclic group or the double dihedral group), or as being a quasi-Hopf algebra with radical of codimension two. In this way we uncover new classes of basic quasi-Hopf algebras of even dimension, as well as new classes of tensor categories.
2025-08-01
{'id': '2508.00597', 'title': 'Biproduct Quasi-Hopf Algebras of Rank 2', 'abstract': 'Inspired by the work of Radford, for $H$ an arbitrary quasi-Hopf algebra we\ndescribe all the Hopf algebras of dimension $2$ within the braided category of\nleft Yetter-Drinfeld modules over $H$ and determine the biproduct quasi-Hopf\nalgebras defined by them. Classes of such biproduct quasi-Hopf algebras are\nobtained by taking $H$ as the Hopf algebra of functions on a group $G$, endowed\nwith the quasi-Hopf algebra structure provided by a non-trivial $3$-cocycle on\n$G$ (especially when $G$ is a finite cyclic group or the double dihedral\ngroup), or as being a quasi-Hopf algebra with radical of codimension two. In\nthis way we uncover new classes of basic quasi-Hopf algebras of even dimension,\nas well as new classes of tensor categories.', 'authors': ['Daniel Bulacu', 'Matteo Misurati'], 'published': '2025-08-01 12:52:10+00:00', 'categories': ['math.QA', 'math.RA', '16W30'], 'url': 'http://arxiv.org/abs/2508.00597v1'}
242
2508.00387
2,508.00387
null
STF: Shallow-Level Temporal Feedback to Enhance Spiking Transformers
Transformer-based Spiking Neural Networks (SNNs) suffer from a great performance gap compared to floating-point Artificial Neural Networks (ANNs) due to the binary nature of spike trains. Recent efforts have introduced deep-level feedback loops to transmit high-level semantic information to narrow this gap. However, these designs often span multiple deep layers, resulting in costly feature transformations, higher parameter overhead, increased energy consumption, and longer inference latency. To address this issue, we propose Shallow-level Temporal Feedback (STF), a lightweight plug-and-play module for the encoding layer, which consists of Temporal-Spatial Position Embedding (TSPE) and Temporal Feedback (TF). Extensive experiments show that STF consistently improves performance across various Transformer-based SNN backbones on static datasets, including CIFAR-10, CIFAR-100, and ImageNet-1K, under different spike timestep settings. Further analysis reveals that STF enhances the diversity of spike patterns, which is key to performance gain. Moreover, evaluations on adversarial robustness and temporal sensitivity confirm that STF outperforms direct coding and its variants, highlighting its potential as a new spike encoding scheme for static scenarios. Our code will be released upon acceptance.
2025-08-01
{'id': '2508.00387', 'title': 'STF: Shallow-Level Temporal Feedback to Enhance Spiking Transformers', 'abstract': 'Transformer-based Spiking Neural Networks (SNNs) suffer from a great\nperformance gap compared to floating-point \\mbox{Artificial} Neural Networks\n(ANNs) due to the binary nature of spike trains. Recent efforts have introduced\ndeep-level feedback loops to transmit high-level semantic information to narrow\nthis gap. However, these designs often span \\mbox{multiple} deep layers,\nresulting in costly feature transformations, higher parameter overhead,\nincreased energy consumption, and longer inference latency. To address this\nissue, we propose Shallow-level Temporal Feedback (STF), a lightweight\nplug-and-play module for the encoding layer, which consists of Temporal-Spatial\nPosition Embedding (TSPE) and Temporal Feedback (TF). Extensive experiments\nshow that STF consistently improves performance across various\nTransformer-based SNN backbones on static datasets, including CIFAR-10,\nCIFAR-100, and ImageNet-1K, under different spike timestep settings. Further\nanalysis reveals that STF enhances the diversity of spike patterns, which is\nkey to performance gain. Moreover, evaluations on adversarial robustness and\ntemporal sensitivity confirm that STF outperforms direct coding and its\nvariants, highlighting its potential as a new spike encoding scheme for static\nscenarios. Our code will be released upon acceptance.', 'authors': ['Zeqi Zheng', 'Zizheng Zhu', 'Yingchao Yu', 'Yanchen Huang', 'Changze Lv', 'Junfeng Tang', 'Zhaofei Yu', 'Yaochu Jin'], 'published': '2025-08-01 07:30:59+00:00', 'categories': ['cs.NE', 'cs.CV'], 'url': 'http://arxiv.org/abs/2508.00387v2'}
243
2508.00760
2,508.0076
null
MMBERT: Scaled Mixture-of-Experts Multimodal BERT for Robust Chinese Hate Speech Detection under Cloaking Perturbations
Hate speech detection on Chinese social networks presents distinct challenges, particularly due to the widespread use of cloaking techniques designed to evade conventional text-based detection systems. Although large language models (LLMs) have recently improved hate speech detection capabilities, the majority of existing work has concentrated on English datasets, with limited attention given to multimodal strategies in the Chinese context. In this study, we propose MMBERT, a novel BERT-based multimodal framework that integrates textual, speech, and visual modalities through a Mixture-of-Experts (MoE) architecture. To address the instability associated with directly integrating MoE into BERT-based models, we develop a progressive three-stage training paradigm. MMBERT incorporates modality-specific experts, a shared self-attention mechanism, and a router-based expert allocation strategy to enhance robustness against adversarial perturbations. Empirical results in several Chinese hate speech datasets show that MMBERT significantly surpasses fine-tuned BERT-based encoder models, fine-tuned LLMs, and LLMs utilizing in-context learning approaches.
2025-08-01
{'id': '2508.00760', 'title': 'MMBERT: Scaled Mixture-of-Experts Multimodal BERT for Robust Chinese Hate Speech Detection under Cloaking Perturbations', 'abstract': 'Hate speech detection on Chinese social networks presents distinct\nchallenges, particularly due to the widespread use of cloaking techniques\ndesigned to evade conventional text-based detection systems. Although large\nlanguage models (LLMs) have recently improved hate speech detection\ncapabilities, the majority of existing work has concentrated on English\ndatasets, with limited attention given to multimodal strategies in the Chinese\ncontext. In this study, we propose MMBERT, a novel BERT-based multimodal\nframework that integrates textual, speech, and visual modalities through a\nMixture-of-Experts (MoE) architecture. To address the instability associated\nwith directly integrating MoE into BERT-based models, we develop a progressive\nthree-stage training paradigm. MMBERT incorporates modality-specific experts, a\nshared self-attention mechanism, and a router-based expert allocation strategy\nto enhance robustness against adversarial perturbations. Empirical results in\nseveral Chinese hate speech datasets show that MMBERT significantly surpasses\nfine-tuned BERT-based encoder models, fine-tuned LLMs, and LLMs utilizing\nin-context learning approaches.', 'authors': ['Qiyao Xue', 'Yuchen Dou', 'Ryan Shi', 'Xiang Lorraine Li', 'Wei Gao'], 'published': '2025-08-01 16:34:57+00:00', 'categories': ['cs.CL', 'cs.AI'], 'url': 'http://arxiv.org/abs/2508.00760v1'}
244
2508.00649
2,508.00649
null
Revisiting Adversarial Patch Defenses on Object Detectors: Unified Evaluation, Large-Scale Dataset, and New Insights
Developing reliable defenses against patch attacks on object detectors has attracted increasing interest. However, we identify that existing defense evaluations lack a unified and comprehensive framework, resulting in inconsistent and incomplete assessments of current methods. To address this issue, we revisit 11 representative defenses and present the first patch defense benchmark, involving 2 attack goals, 13 patch attacks, 11 object detectors, and 4 diverse metrics. This leads to the large-scale adversarial patch dataset with 94 types of patches and 94,000 images. Our comprehensive analyses reveal new insights: (1) The difficulty in defending against naturalistic patches lies in the data distribution, rather than the commonly believed high frequencies. Our new dataset with diverse patch distributions can be used to improve existing defenses by 15.09% [email protected]. (2) The average precision of the attacked object, rather than the commonly pursued patch detection accuracy, shows high consistency with defense performance. (3) Adaptive attacks can substantially bypass existing defenses, and defenses with complex/stochastic models or universal patch properties are relatively robust. We hope that our analyses will serve as guidance on properly evaluating patch attacks/defenses and advancing their design. Code and dataset are available at https://github.com/Gandolfczjh/APDE, where we will keep integrating new attacks/defenses.
2025-08-01
{'id': '2508.00649', 'title': 'Revisiting Adversarial Patch Defenses on Object Detectors: Unified Evaluation, Large-Scale Dataset, and New Insights', 'abstract': 'Developing reliable defenses against patch attacks on object detectors has\nattracted increasing interest. However, we identify that existing defense\nevaluations lack a unified and comprehensive framework, resulting in\ninconsistent and incomplete assessments of current methods. To address this\nissue, we revisit 11 representative defenses and present the first patch\ndefense benchmark, involving 2 attack goals, 13 patch attacks, 11 object\ndetectors, and 4 diverse metrics. This leads to the large-scale adversarial\npatch dataset with 94 types of patches and 94,000 images. Our comprehensive\nanalyses reveal new insights: (1) The difficulty in defending against\nnaturalistic patches lies in the data distribution, rather than the commonly\nbelieved high frequencies. Our new dataset with diverse patch distributions can\nbe used to improve existing defenses by 15.09% [email protected]. (2) The average\nprecision of the attacked object, rather than the commonly pursued patch\ndetection accuracy, shows high consistency with defense performance. (3)\nAdaptive attacks can substantially bypass existing defenses, and defenses with\ncomplex/stochastic models or universal patch properties are relatively robust.\nWe hope that our analyses will serve as guidance on properly evaluating patch\nattacks/defenses and advancing their design. Code and dataset are available at\nhttps://github.com/Gandolfczjh/APDE, where we will keep integrating new\nattacks/defenses.', 'authors': ['Junhao Zheng', 'Jiahao Sun', 'Chenhao Lin', 'Zhengyu Zhao', 'Chen Ma', 'Chong Zhang', 'Cong Wang', 'Qian Wang', 'Chao Shen'], 'published': '2025-08-01 14:06:20+00:00', 'categories': ['cs.CV', 'cs.CR'], 'url': 'http://arxiv.org/abs/2508.00649v2'}
245
2508.00511
2,508.00511
null
On the regularity of almost stable relations
We develop a general theory of local stability up to belonging to an ideal (e.g. having measure zero). From a model-theoretic perspective, we prove a stationarity principle for almost stable formulas in this sense, and build a topological space of partial types whose Cantor-Bendixson rank is finite. The interaction of this space with Keisler measures and definable groups yields, on the one hand, a regularity lemma for infinite graphs where the edge relation is almost stable, and, on the other hand, the existence of definable stabilizer subgroups. As an application, we prove a finite graph regularity lemma and an arithmetic regularity lemma for almost stable relations in arbitrary finite groups.
2025-08-01
{'id': '2508.00511', 'title': 'On the regularity of almost stable relations', 'abstract': 'We develop a general theory of local stability up to belonging to an ideal\n(e.g. having measure zero). From a model-theoretic perspective, we prove a\nstationarity principle for almost stable formulas in this sense, and build a\ntopological space of partial types whose Cantor-Bendixson rank is finite. The\ninteraction of this space with Keisler measures and definable groups yields, on\nthe one hand, a regularity lemma for infinite graphs where the edge relation is\nalmost stable, and, on the other hand, the existence of definable stabilizer\nsubgroups. As an application, we prove a finite graph regularity lemma and an\narithmetic regularity lemma for almost stable relations in arbitrary finite\ngroups.', 'authors': ['Marcos Girón'], 'published': '2025-08-01 10:43:41+00:00', 'categories': ['math.LO'], 'url': 'http://arxiv.org/abs/2508.00511v1'}
246
2508.00555
2,508.00555
null
Activation-Guided Local Editing for Jailbreaking Attacks
Jailbreaking is an essential adversarial technique for red-teaming these models to uncover and patch security flaws. However, existing jailbreak methods face significant drawbacks. Token-level jailbreak attacks often produce incoherent or unreadable inputs and exhibit poor transferability, while prompt-level attacks lack scalability and rely heavily on manual effort and human ingenuity. We propose a concise and effective two-stage framework that combines the advantages of these approaches. The first stage performs a scenario-based generation of context and rephrases the original malicious query to obscure its harmful intent. The second stage then utilizes information from the model's hidden states to guide fine-grained edits, effectively steering the model's internal representation of the input from a malicious toward a benign one. Extensive experiments demonstrate that this method achieves state-of-the-art Attack Success Rate, with gains of up to 37.74% over the strongest baseline, and exhibits excellent transferability to black-box models. Our analysis further demonstrates that AGILE maintains substantial effectiveness against prominent defense mechanisms, highlighting the limitations of current safeguards and providing valuable insights for future defense development. Our code is available at https://github.com/yunsaijc/AGILE.
2025-08-01
{'id': '2508.00555', 'title': 'Activation-Guided Local Editing for Jailbreaking Attacks', 'abstract': "Jailbreaking is an essential adversarial technique for red-teaming these\nmodels to uncover and patch security flaws. However, existing jailbreak methods\nface significant drawbacks. Token-level jailbreak attacks often produce\nincoherent or unreadable inputs and exhibit poor transferability, while\nprompt-level attacks lack scalability and rely heavily on manual effort and\nhuman ingenuity. We propose a concise and effective two-stage framework that\ncombines the advantages of these approaches. The first stage performs a\nscenario-based generation of context and rephrases the original malicious query\nto obscure its harmful intent. The second stage then utilizes information from\nthe model's hidden states to guide fine-grained edits, effectively steering the\nmodel's internal representation of the input from a malicious toward a benign\none. Extensive experiments demonstrate that this method achieves\nstate-of-the-art Attack Success Rate, with gains of up to 37.74% over the\nstrongest baseline, and exhibits excellent transferability to black-box models.\nOur analysis further demonstrates that AGILE maintains substantial\neffectiveness against prominent defense mechanisms, highlighting the\nlimitations of current safeguards and providing valuable insights for future\ndefense development. Our code is available at\nhttps://github.com/yunsaijc/AGILE.", 'authors': ['Jiecong Wang', 'Haoran Li', 'Hao Peng', 'Ziqian Zeng', 'Zihao Wang', 'Haohua Du', 'Zhengtao Yu'], 'published': '2025-08-01 11:52:24+00:00', 'categories': ['cs.CR', 'cs.AI', 'cs.CL'], 'url': 'http://arxiv.org/abs/2508.00555v1'}
247
2508.00226
2,508.00226
null
Neutrino magnetic moment in the doublet-singlet Leptoquark model
The neutrino transition magnetic moment $__$ is studied in a simple extension of the Standard Model. This extension incorporates two scalar Leptoquarks $S_1$ and $R_2$ with quantum numbers $(3,1,1/3)$ and $(3,2,1/6)$ respectively. It is found that these Leptoquarks generate a sizable transition magnetic moment, particularly when the quark bottom is running in the loop. For our analysis of the parameter space, we include the latest measurement of the muon magnetic moment and combine it with the experimental constraint on the branching ratio Br$()$. We found that, despite the recent agreement on the $(g-2)_$ value, large values for Leptoquark Yukawa couplings are allowed due to a degeneracy in the parameters. Additionally, we explore how the Leptoquark model address the anomalies observed in the ratios of semileptonic $B$ mesons decays, $R_D^(*)$. We determine that the restrictions derived from our analysis are consistent with the most recent experimental limits reported by the XENONnT and LUX-ZEPLIN collaborations. This conclusion is based on our evaluation of the transition magnetic moment from muon neutrino to tau neutrino, focusing on the allowed region for the Leptoquark Yukawa couplings.
2025-08-01
{'id': '2508.00226', 'title': 'Neutrino magnetic moment in the doublet-singlet Leptoquark model', 'abstract': 'The neutrino transition magnetic moment $\\mu_{\\nu_{\\alpha \\beta}}$ is studied\nin a simple extension of the Standard Model. This extension incorporates two\nscalar Leptoquarks $S_1$ and $\\widetilde{R}_2$ with quantum numbers\n$(\\bar{3},1,1/3)$ and $(3,2,1/6)$ respectively. It is found that these\nLeptoquarks generate a sizable transition magnetic moment, particularly when\nthe quark bottom is running in the loop. For our analysis of the parameter\nspace, we include the latest measurement of the muon magnetic moment and\ncombine it with the experimental constraint on the branching ratio Br$(\\tau \\to\n\\mu \\gamma)$. We found that, despite the recent agreement on the $(g-2)_\\mu$\nvalue, large values for Leptoquark Yukawa couplings are allowed due to a\ndegeneracy in the parameters. Additionally, we explore how the Leptoquark model\naddress the anomalies observed in the ratios of semileptonic $B$ mesons decays,\n$R_{D^{(*)}}$. We determine that the restrictions derived from our analysis are\nconsistent with the most recent experimental limits reported by the XENONnT and\nLUX-ZEPLIN collaborations. This conclusion is based on our evaluation of the\ntransition magnetic moment from muon neutrino to tau neutrino, focusing on the\nallowed region for the Leptoquark Yukawa couplings.', 'authors': ['Ricardo Sánchez-Vélez'], 'published': '2025-08-01 00:11:40+00:00', 'categories': ['hep-ph'], 'url': 'http://arxiv.org/abs/2508.00226v1'}
248
2508.00325
2,508.00325
null
PnP-DA: Towards Principled Plug-and-Play Integration of Variational Data Assimilation and Generative Models
Earth system modeling presents a fundamental challenge in scientific computing: capturing complex, multiscale nonlinear dynamics in computationally efficient models while minimizing forecast errors caused by necessary simplifications. Even the most powerful AI- or physics-based forecast system suffer from gradual error accumulation. Data assimilation (DA) aims to mitigate these errors by optimally blending (noisy) observations with prior model forecasts, but conventional variational methods often assume Gaussian error statistics that fail to capture the true, non-Gaussian behavior of chaotic dynamical systems. We propose PnP-DA, a Plug-and-Play algorithm that alternates (1) a lightweight, gradient-based analysis update (using a Mahalanobis-distance misfit on new observations) with (2) a single forward pass through a pretrained generative prior conditioned on the background forecast via a conditional Wasserstein coupling. This strategy relaxes restrictive statistical assumptions and leverages rich historical data without requiring an explicit regularization functional, and it also avoids the need to backpropagate gradients through the complex neural network that encodes the prior during assimilation cycles. Experiments on standard chaotic testbeds demonstrate that this strategy consistently reduces forecast errors across a range of observation sparsities and noise levels, outperforming classical variational methods.
2025-08-01
{'id': '2508.00325', 'title': 'PnP-DA: Towards Principled Plug-and-Play Integration of Variational Data Assimilation and Generative Models', 'abstract': 'Earth system modeling presents a fundamental challenge in scientific\ncomputing: capturing complex, multiscale nonlinear dynamics in computationally\nefficient models while minimizing forecast errors caused by necessary\nsimplifications. Even the most powerful AI- or physics-based forecast system\nsuffer from gradual error accumulation. Data assimilation (DA) aims to mitigate\nthese errors by optimally blending (noisy) observations with prior model\nforecasts, but conventional variational methods often assume Gaussian error\nstatistics that fail to capture the true, non-Gaussian behavior of chaotic\ndynamical systems. We propose PnP-DA, a Plug-and-Play algorithm that alternates\n(1) a lightweight, gradient-based analysis update (using a Mahalanobis-distance\nmisfit on new observations) with (2) a single forward pass through a pretrained\ngenerative prior conditioned on the background forecast via a conditional\nWasserstein coupling. This strategy relaxes restrictive statistical assumptions\nand leverages rich historical data without requiring an explicit regularization\nfunctional, and it also avoids the need to backpropagate gradients through the\ncomplex neural network that encodes the prior during assimilation cycles.\nExperiments on standard chaotic testbeds demonstrate that this strategy\nconsistently reduces forecast errors across a range of observation sparsities\nand noise levels, outperforming classical variational methods.', 'authors': ['Yongquan Qu', 'Matthieu Blanke', 'Sara Shamekh', 'Pierre Gentine'], 'published': '2025-08-01 05:19:19+00:00', 'categories': ['cs.LG', 'physics.comp-ph'], 'url': 'http://arxiv.org/abs/2508.00325v1'}
249
2508.00230
2,508.0023
null
Towards Higher Effective Rank in Parameter-efficient Fine-tuning using Khatri--Rao Product
Parameter-efficient fine-tuning (PEFT) has become a standard approach for adapting large pre-trained models. Amongst PEFT methods, low-rank adaptation (LoRA) has achieved notable success. However, recent studies have highlighted its limitations compared against full-rank alternatives, particularly when applied to multimodal and large language models. In this work, we present a quantitative comparison amongst full-rank and low-rank PEFT methods using a synthetic matrix approximation benchmark with controlled spectral properties. Our results confirm that LoRA struggles to approximate matrices with relatively flat spectrums or high frequency components -- signs of high effective ranks. To this end, we introduce KRAdapter, a novel PEFT algorithm that leverages the Khatri-Rao product to produce weight updates, which, by construction, tends to produce matrix product with a high effective rank. We demonstrate performance gains with KRAdapter on vision-language models up to 1B parameters and on large language models up to 8B parameters, particularly on unseen common-sense reasoning tasks. In addition, KRAdapter maintains the memory and compute efficiency of LoRA, making it a practical and robust alternative to fine-tune billion-scale parameter models.
2025-08-01
{'id': '2508.00230', 'title': 'Towards Higher Effective Rank in Parameter-efficient Fine-tuning using Khatri--Rao Product', 'abstract': 'Parameter-efficient fine-tuning (PEFT) has become a standard approach for\nadapting large pre-trained models. Amongst PEFT methods, low-rank adaptation\n(LoRA) has achieved notable success. However, recent studies have highlighted\nits limitations compared against full-rank alternatives, particularly when\napplied to multimodal and large language models. In this work, we present a\nquantitative comparison amongst full-rank and low-rank PEFT methods using a\nsynthetic matrix approximation benchmark with controlled spectral properties.\nOur results confirm that LoRA struggles to approximate matrices with relatively\nflat spectrums or high frequency components -- signs of high effective ranks.\nTo this end, we introduce KRAdapter, a novel PEFT algorithm that leverages the\nKhatri-Rao product to produce weight updates, which, by construction, tends to\nproduce matrix product with a high effective rank. We demonstrate performance\ngains with KRAdapter on vision-language models up to 1B parameters and on large\nlanguage models up to 8B parameters, particularly on unseen common-sense\nreasoning tasks. In addition, KRAdapter maintains the memory and compute\nefficiency of LoRA, making it a practical and robust alternative to fine-tune\nbillion-scale parameter models.', 'authors': ['Paul Albert', 'Frederic Z. Zhang', 'Hemanth Saratchandran', 'Anton van den Hengel', 'Ehsan Abbasnejad'], 'published': '2025-08-01 00:29:13+00:00', 'categories': ['cs.LG', 'cs.CL', 'cs.CV'], 'url': 'http://arxiv.org/abs/2508.00230v1'}
250
2508.00755
2,508.00755
null
AI-Driven Collaborative Satellite Object Detection for Space Sustainability
The growing density of satellites in low-Earth orbit (LEO) presents serious challenges to space sustainability, primarily due to the increased risk of in-orbit collisions. Traditional ground-based tracking systems are constrained by latency and coverage limitations, underscoring the need for onboard, vision-based space object detection (SOD) capabilities. In this paper, we propose a novel satellite clustering framework that enables the collaborative execution of deep learning (DL)-based SOD tasks across multiple satellites. To support this approach, we construct a high-fidelity dataset simulating imaging scenarios for clustered satellite formations. A distance-aware viewpoint selection strategy is introduced to optimize detection performance, and recent DL models are used for evaluation. Experimental results show that the clustering-based method achieves competitive detection accuracy compared to single-satellite and existing approaches, while maintaining a low size, weight, and power (SWaP) footprint. These findings underscore the potential of distributed, AI-enabled in-orbit systems to enhance space situational awareness and contribute to long-term space sustainability.
2025-08-01
{'id': '2508.00755', 'title': 'AI-Driven Collaborative Satellite Object Detection for Space Sustainability', 'abstract': 'The growing density of satellites in low-Earth orbit (LEO) presents serious\nchallenges to space sustainability, primarily due to the increased risk of\nin-orbit collisions. Traditional ground-based tracking systems are constrained\nby latency and coverage limitations, underscoring the need for onboard,\nvision-based space object detection (SOD) capabilities. In this paper, we\npropose a novel satellite clustering framework that enables the collaborative\nexecution of deep learning (DL)-based SOD tasks across multiple satellites. To\nsupport this approach, we construct a high-fidelity dataset simulating imaging\nscenarios for clustered satellite formations. A distance-aware viewpoint\nselection strategy is introduced to optimize detection performance, and recent\nDL models are used for evaluation. Experimental results show that the\nclustering-based method achieves competitive detection accuracy compared to\nsingle-satellite and existing approaches, while maintaining a low size, weight,\nand power (SWaP) footprint. These findings underscore the potential of\ndistributed, AI-enabled in-orbit systems to enhance space situational awareness\nand contribute to long-term space sustainability.', 'authors': ['Peng Hu', 'Wenxuan Zhang'], 'published': '2025-08-01 16:31:55+00:00', 'categories': ['eess.IV', 'cs.CV'], 'url': 'http://arxiv.org/abs/2508.00755v1'}
251
2508.00676
2,508.00676
null
AlF-AlF sticking time and prospects for ultracold dimers
We report on the sticking time of the AlF dimer in the ultracold regime. We employ a full-dimensional potential energy surface for AlF-AlF, constructed using a machine learning approach [X. Liu et al., J. Chem. Phys. 159, 144103 (2023)], to compute the density of states using a semi-classical counting method. Next, using the Rice-Ramsperger-Kassel-Marcus (RRKM) theory, we determine a sticking time of 216.3 ns, which is shorter than that of other previously reported dimers. We explain these results in light of the ratio of the dissociation energy of the complex to the dissociation energy of the molecule, yielding a computationally inexpensive scheme to estimate the sticking time of collisional complexes.
2025-08-01
{'id': '2508.00676', 'title': 'AlF-AlF sticking time and prospects for ultracold dimers', 'abstract': 'We report on the sticking time of the AlF dimer in the ultracold regime. We\nemploy a full-dimensional potential energy surface for AlF-AlF, constructed\nusing a machine learning approach [X. Liu et al., J. Chem. Phys. 159, 144103\n(2023)], to compute the density of states using a semi-classical counting\nmethod. Next, using the Rice-Ramsperger-Kassel-Marcus (RRKM) theory, we\ndetermine a sticking time of 216.3 ns, which is shorter than that of other\npreviously reported dimers. We explain these results in light of the ratio of\nthe dissociation energy of the complex to the dissociation energy of the\nmolecule, yielding a computationally inexpensive scheme to estimate the\nsticking time of collisional complexes.', 'authors': ['Mahmoud A. E. Ibrahim', 'Mateo Londoño', 'Jesús Pérez-Ríos'], 'published': '2025-08-01 14:48:51+00:00', 'categories': ['physics.atm-clus'], 'url': 'http://arxiv.org/abs/2508.00676v1'}
252
2508.00337
2,508.00337
null
Nonlocal free boundary minimal surfaces
We introduce the nonlocal analogue of the classical free boundary minimal hypersurfaces in an open domain $$ of $R^n$ as the (boundaries of) critical points of the fractional perimeter $Per_s(,\,)$ with respect to inner variations leaving $$ invariant. We deduce the Euler-Lagrange equations and prove a few surprising features, such as the existence of critical points without boundary and a strong volume constraint in $$ for unbounded hypersurfaces. Moreover, we investigate stickiness properties and regularity across the boundary.
2025-08-01
{'id': '2508.00337', 'title': 'Nonlocal free boundary minimal surfaces', 'abstract': 'We introduce the nonlocal analogue of the classical free boundary minimal\nhypersurfaces in an open domain $\\Omega$ of $\\mathbb{R}^n$ as the (boundaries\nof) critical points of the fractional perimeter\n$\\operatorname{Per}_s(\\cdot,\\,\\Omega )$ with respect to inner variations\nleaving $\\Omega$ invariant. We deduce the Euler-Lagrange equations and prove a\nfew surprising features, such as the existence of critical points without\nboundary and a strong volume constraint in $\\Omega$ for unbounded\nhypersurfaces. Moreover, we investigate stickiness properties and regularity\nacross the boundary.', 'authors': ['Marco Badran', 'Serena Dipierro', 'Enrico Valdinoci'], 'published': '2025-08-01 05:55:25+00:00', 'categories': ['math.AP', 'math.DG', '49Q05, 35R11, 35R35, 49Q10'], 'url': 'http://arxiv.org/abs/2508.00337v1'}
253
2508.00354
2,508.00354
null
Omni-Scan: Creating Visually-Accurate Digital Twin Object Models Using a Bimanual Robot with Handover and Gaussian Splat Merging
3D Gaussian Splats (3DGSs) are 3D object models derived from multi-view images. Such "digital twins" are useful for simulations, virtual reality, marketing, robot policy fine-tuning, and part inspection. 3D object scanning usually requires multi-camera arrays, precise laser scanners, or robot wrist-mounted cameras, which have restricted workspaces. We propose Omni-Scan, a pipeline for producing high-quality 3D Gaussian Splat models using a bi-manual robot that grasps an object with one gripper and rotates the object with respect to a stationary camera. The object is then re-grasped by a second gripper to expose surfaces that were occluded by the first gripper. We present the Omni-Scan robot pipeline using DepthAny-thing, Segment Anything, as well as RAFT optical flow models to identify and isolate objects held by a robot gripper while removing the gripper and the background. We then modify the 3DGS training pipeline to support concatenated datasets with gripper occlusion, producing an omni-directional (360 degree view) model of the object. We apply Omni-Scan to part defect inspection, finding that it can identify visual or geometric defects in 12 different industrial and household objects with an average accuracy of 83%. Interactive videos of Omni-Scan 3DGS models can be found at https://berkeleyautomation.github.io/omni-scan/
2025-08-01
{'id': '2508.00354', 'title': 'Omni-Scan: Creating Visually-Accurate Digital Twin Object Models Using a Bimanual Robot with Handover and Gaussian Splat Merging', 'abstract': '3D Gaussian Splats (3DGSs) are 3D object models derived from multi-view\nimages. Such "digital twins" are useful for simulations, virtual reality,\nmarketing, robot policy fine-tuning, and part inspection. 3D object scanning\nusually requires multi-camera arrays, precise laser scanners, or robot\nwrist-mounted cameras, which have restricted workspaces. We propose Omni-Scan,\na pipeline for producing high-quality 3D Gaussian Splat models using a\nbi-manual robot that grasps an object with one gripper and rotates the object\nwith respect to a stationary camera. The object is then re-grasped by a second\ngripper to expose surfaces that were occluded by the first gripper. We present\nthe Omni-Scan robot pipeline using DepthAny-thing, Segment Anything, as well as\nRAFT optical flow models to identify and isolate objects held by a robot\ngripper while removing the gripper and the background. We then modify the 3DGS\ntraining pipeline to support concatenated datasets with gripper occlusion,\nproducing an omni-directional (360 degree view) model of the object. We apply\nOmni-Scan to part defect inspection, finding that it can identify visual or\ngeometric defects in 12 different industrial and household objects with an\naverage accuracy of 83%. Interactive videos of Omni-Scan 3DGS models can be\nfound at https://berkeleyautomation.github.io/omni-scan/', 'authors': ['Tianshuang Qiu', 'Zehan Ma', 'Karim El-Refai', 'Hiya Shah', 'Chung Min Kim', 'Justin Kerr', 'Ken Goldberg'], 'published': '2025-08-01 06:36:19+00:00', 'categories': ['cs.RO', 'cs.CV'], 'url': 'http://arxiv.org/abs/2508.00354v1'}
254
2508.00622
2,508.00622
null
SwarnRaft: Leveraging Consensus for Robust Drone Swarm Coordination in GNSS-Degraded Environments
Unmanned aerial vehicle (UAV) swarms are increasingly used in critical applications such as aerial mapping, environmental monitoring, and autonomous delivery. However, the reliability of these systems is highly dependent on uninterrupted access to the Global Navigation Satellite Systems (GNSS) signals, which can be disrupted in real-world scenarios due to interference, environmental conditions, or adversarial attacks, causing disorientation, collision risks, and mission failure. This paper proposes SwarnRaft, a blockchain-inspired positioning and consensus framework for maintaining coordination and data integrity in UAV swarms operating under GNSS-denied conditions. SwarnRaft leverages the Raft consensus algorithm to enable distributed drones (nodes) to agree on state updates such as location and heading, even in the absence of GNSS signals for one or more nodes. In our prototype, each node uses GNSS and local sensing, and communicates over WiFi in a simulated swarm. Upon signal loss, consensus is used to reconstruct or verify the position of the failed node based on its last known state and trajectory. Our system demonstrates robustness in maintaining swarm coherence and fault tolerance through a lightweight, scalable communication model. This work offers a practical and secure foundation for decentralized drone operation in unpredictable environments.
2025-08-01
{'id': '2508.00622', 'title': 'SwarnRaft: Leveraging Consensus for Robust Drone Swarm Coordination in GNSS-Degraded Environments', 'abstract': 'Unmanned aerial vehicle (UAV) swarms are increasingly used in critical\napplications such as aerial mapping, environmental monitoring, and autonomous\ndelivery. However, the reliability of these systems is highly dependent on\nuninterrupted access to the Global Navigation Satellite Systems (GNSS) signals,\nwhich can be disrupted in real-world scenarios due to interference,\nenvironmental conditions, or adversarial attacks, causing disorientation,\ncollision risks, and mission failure. This paper proposes SwarnRaft, a\nblockchain-inspired positioning and consensus framework for maintaining\ncoordination and data integrity in UAV swarms operating under GNSS-denied\nconditions. SwarnRaft leverages the Raft consensus algorithm to enable\ndistributed drones (nodes) to agree on state updates such as location and\nheading, even in the absence of GNSS signals for one or more nodes. In our\nprototype, each node uses GNSS and local sensing, and communicates over WiFi in\na simulated swarm. Upon signal loss, consensus is used to reconstruct or verify\nthe position of the failed node based on its last known state and trajectory.\nOur system demonstrates robustness in maintaining swarm coherence and fault\ntolerance through a lightweight, scalable communication model. This work offers\na practical and secure foundation for decentralized drone operation in\nunpredictable environments.', 'authors': ['Kapel Dev', 'Yash Madhwal', 'Sofia Shevelo', 'Pavel Osinenko', 'Yury Yanovich'], 'published': '2025-08-01 13:33:59+00:00', 'categories': ['cs.DC'], 'url': 'http://arxiv.org/abs/2508.00622v1'}
255
2508.00748
2,508.00748
null
Is It Really You? Exploring Biometric Verification Scenarios in Photorealistic Talking-Head Avatar Videos
Photorealistic talking-head avatars are becoming increasingly common in virtual meetings, gaming, and social platforms. These avatars allow for more immersive communication, but they also introduce serious security risks. One emerging threat is impersonation: an attacker can steal a user's avatar, preserving his appearance and voice, making it nearly impossible to detect its fraudulent usage by sight or sound alone. In this paper, we explore the challenge of biometric verification in such avatar-mediated scenarios. Our main question is whether an individual's facial motion patterns can serve as reliable behavioral biometrics to verify their identity when the avatar's visual appearance is a facsimile of its owner. To answer this question, we introduce a new dataset of realistic avatar videos created using a state-of-the-art one-shot avatar generation model, GAGAvatar, with genuine and impostor avatar videos. We also propose a lightweight, explainable spatio-temporal Graph Convolutional Network architecture with temporal attention pooling, that uses only facial landmarks to model dynamic facial gestures. Experimental results demonstrate that facial motion cues enable meaningful identity verification with AUC values approaching 80%. The proposed benchmark and biometric system are available for the research community in order to bring attention to the urgent need for more advanced behavioral biometric defenses in avatar-based communication systems.
2025-08-01
{'id': '2508.00748', 'title': 'Is It Really You? Exploring Biometric Verification Scenarios in Photorealistic Talking-Head Avatar Videos', 'abstract': "Photorealistic talking-head avatars are becoming increasingly common in\nvirtual meetings, gaming, and social platforms. These avatars allow for more\nimmersive communication, but they also introduce serious security risks. One\nemerging threat is impersonation: an attacker can steal a user's avatar,\npreserving his appearance and voice, making it nearly impossible to detect its\nfraudulent usage by sight or sound alone. In this paper, we explore the\nchallenge of biometric verification in such avatar-mediated scenarios. Our main\nquestion is whether an individual's facial motion patterns can serve as\nreliable behavioral biometrics to verify their identity when the avatar's\nvisual appearance is a facsimile of its owner. To answer this question, we\nintroduce a new dataset of realistic avatar videos created using a\nstate-of-the-art one-shot avatar generation model, GAGAvatar, with genuine and\nimpostor avatar videos. We also propose a lightweight, explainable\nspatio-temporal Graph Convolutional Network architecture with temporal\nattention pooling, that uses only facial landmarks to model dynamic facial\ngestures. Experimental results demonstrate that facial motion cues enable\nmeaningful identity verification with AUC values approaching 80%. The proposed\nbenchmark and biometric system are available for the research community in\norder to bring attention to the urgent need for more advanced behavioral\nbiometric defenses in avatar-based communication systems.", 'authors': ['Laura Pedrouzo-Rodriguez', 'Pedro Delgado-DeRobles', 'Luis F. Gomez', 'Ruben Tolosana', 'Ruben Vera-Rodriguez', 'Aythami Morales', 'Julian Fierrez'], 'published': '2025-08-01 16:23:27+00:00', 'categories': ['cs.CV', 'cs.AI', 'cs.CR', 'cs.MM'], 'url': 'http://arxiv.org/abs/2508.00748v2'}
256
2508.00626
2,508.00626
null
Deep Learning-Based Rate-Adaptive CSI Feedback for Wideband XL-MIMO Systems in the Near-Field Domain
Accurate and efficient channel state information (CSI) feedback is crucial for unlocking the substantial spectral efficiency gains of extremely large-scale MIMO (XL-MIMO) systems in future 6G networks. However, the combination of near-field spherical wave propagation and frequency-dependent beam split effects in wideband scenarios poses significant challenges for CSI representation and compression. This paper proposes WideNLNet-CA, a rate-adaptive deep learning framework designed to enable efficient CSI feedback in wideband near-field XL-MIMO systems. WideNLNet-CA introduces a lightweight encoder-decoder architecture with multi-stage downsampling and upsampling, incorporating computationally efficient residual blocks to capture complex multi-scale channel features with reduced overhead. A novel compression ratio adaptive module with feature importance estimation is introduced to dynamically modulate feature selection based on target compression ratios, enabling flexible adaptation across a wide range of feedback rates using a single model. Evaluation results demonstrate that WideNLNet-CA consistently outperforms existing compressive sensing and deep learning-based works across various compression ratios and bandwidths, while maintaining fast inference and low model storage requirements.
2025-08-01
{'id': '2508.00626', 'title': 'Deep Learning-Based Rate-Adaptive CSI Feedback for Wideband XL-MIMO Systems in the Near-Field Domain', 'abstract': 'Accurate and efficient channel state information (CSI) feedback is crucial\nfor unlocking the substantial spectral efficiency gains of extremely\nlarge-scale MIMO (XL-MIMO) systems in future 6G networks. However, the\ncombination of near-field spherical wave propagation and frequency-dependent\nbeam split effects in wideband scenarios poses significant challenges for CSI\nrepresentation and compression. This paper proposes WideNLNet-CA, a\nrate-adaptive deep learning framework designed to enable efficient CSI feedback\nin wideband near-field XL-MIMO systems. WideNLNet-CA introduces a lightweight\nencoder-decoder architecture with multi-stage downsampling and upsampling,\nincorporating computationally efficient residual blocks to capture complex\nmulti-scale channel features with reduced overhead. A novel compression ratio\nadaptive module with feature importance estimation is introduced to dynamically\nmodulate feature selection based on target compression ratios, enabling\nflexible adaptation across a wide range of feedback rates using a single model.\nEvaluation results demonstrate that WideNLNet-CA consistently outperforms\nexisting compressive sensing and deep learning-based works across various\ncompression ratios and bandwidths, while maintaining fast inference and low\nmodel storage requirements.', 'authors': ['Zhenyu Liu', 'Yi Ma', 'Rahim Tafazolli'], 'published': '2025-08-01 13:38:38+00:00', 'categories': ['cs.IT', 'eess.SP', 'math.IT'], 'url': 'http://arxiv.org/abs/2508.00626v1'}
263
2508.00474
2,508.00474
null
Linear F-manifolds, a duality and the generalized tangent bundle
A linear F-manifold is an F-manifold (E, , e) defined on the total space of a vector bundle : E M for which the multiplication and unit field are linear tensor fields. We develop a systematic treatment of linear F-manifolds. Using an additional suitable connection on M, we define a duality between linear F-manifolds (with and without Euler fields) on E and the total space E^* of the dual vector bundle. Our main examples of linear F-manifolds are the tangent and cotangent prolongation. Motivated by the direct sum of tangent and cotangent prolongation, we define and investigate compatibility conditions between linear F-manifolds and the geometry of the generalized tangent bundle.
2025-08-01
{'id': '2508.00474', 'title': 'Linear F-manifolds, a duality and the generalized tangent bundle', 'abstract': 'A linear F-manifold is an F-manifold (E, \\circ , e) defined on the total\nspace of a vector bundle \\pi : E \\rightarrow M for which the multiplication and\nunit field are linear tensor fields. We develop a systematic treatment of\nlinear F-manifolds. Using an additional suitable connection on M, we define a\nduality between linear F-manifolds (with and without Euler fields) on E and the\ntotal space E^{*} of the dual vector bundle. Our main examples of linear\nF-manifolds are the tangent and cotangent prolongation. Motivated by the direct\nsum of tangent and cotangent prolongation, we define and investigate\ncompatibility conditions between linear F-manifolds and the geometry of the\ngeneralized tangent bundle.', 'authors': ['Liana David'], 'published': '2025-08-01 09:50:21+00:00', 'categories': ['math.DG', 'math-ph', 'math.MP'], 'url': 'http://arxiv.org/abs/2508.00474v1'}
257
2508.00606
2,508.00606
null
New torsion patterns in Khovanov homology
In a previous paper by the authors, we found some patterns in link diagrams that give rise to torsion elements of order two in their Khovanov homology. In this paper we extend these results by providing new torsion patterns. Many of the torsion elements found in this way have the same homological and quantum degrees; we identify a type of submodules of the Khovanov chain complex that allows us to prove that most of these torsion elements living in the same Khovanov module are really different. We use the results of this paper together with those in the previous one to find all the torsion elements in many small twists knots. In addition, we apply them to determine torsion elements in some families of pretzel links, closures of braids with three strands and rational links.
2025-08-01
{'id': '2508.00606', 'title': 'New torsion patterns in Khovanov homology', 'abstract': 'In a previous paper by the authors, we found some patterns in link diagrams\nthat give rise to torsion elements of order two in their Khovanov homology. In\nthis paper we extend these results by providing new torsion patterns. Many of\nthe torsion elements found in this way have the same homological and quantum\ndegrees; we identify a type of submodules of the Khovanov chain complex that\nallows us to prove that most of these torsion elements living in the same\nKhovanov module are really different.\n We use the results of this paper together with those in the previous one to\nfind all the torsion elements in many small twists knots. In addition, we apply\nthem to determine torsion elements in some families of pretzel links, closures\nof braids with three strands and rational links.', 'authors': ['Raquel Díaz', 'Pedro M. G. Manchón'], 'published': '2025-08-01 13:08:48+00:00', 'categories': ['math.GT', '57K10, 57K18'], 'url': 'http://arxiv.org/abs/2508.00606v1'}
258
2508.00442
2,508.00442
null
TopoTTA: Topology-Enhanced Test-Time Adaptation for Tubular Structure Segmentation
Tubular structure segmentation (TSS) is important for various applications, such as hemodynamic analysis and route navigation. Despite significant progress in TSS, domain shifts remain a major challenge, leading to performance degradation in unseen target domains. Unlike other segmentation tasks, TSS is more sensitive to domain shifts, as changes in topological structures can compromise segmentation integrity, and variations in local features distinguishing foreground from background (e.g., texture and contrast) may further disrupt topological continuity. To address these challenges, we propose Topology-enhanced Test-Time Adaptation (TopoTTA), the first test-time adaptation framework designed specifically for TSS. TopoTTA consists of two stages: Stage 1 adapts models to cross-domain topological discrepancies using the proposed Topological Meta Difference Convolutions (TopoMDCs), which enhance topological representation without altering pre-trained parameters; Stage 2 improves topological continuity by a novel Topology Hard sample Generation (TopoHG) strategy and prediction alignment on hard samples with pseudo-labels in the generated pseudo-break regions. Extensive experiments across four scenarios and ten datasets demonstrate TopoTTA's effectiveness in handling topological distribution shifts, achieving an average improvement of 31.81% in clDice. TopoTTA also serves as a plug-and-play TTA solution for CNN-based TSS models.
2025-08-01
{'id': '2508.00442', 'title': 'TopoTTA: Topology-Enhanced Test-Time Adaptation for Tubular Structure Segmentation', 'abstract': "Tubular structure segmentation (TSS) is important for various applications,\nsuch as hemodynamic analysis and route navigation. Despite significant progress\nin TSS, domain shifts remain a major challenge, leading to performance\ndegradation in unseen target domains. Unlike other segmentation tasks, TSS is\nmore sensitive to domain shifts, as changes in topological structures can\ncompromise segmentation integrity, and variations in local features\ndistinguishing foreground from background (e.g., texture and contrast) may\nfurther disrupt topological continuity. To address these challenges, we propose\nTopology-enhanced Test-Time Adaptation (TopoTTA), the first test-time\nadaptation framework designed specifically for TSS. TopoTTA consists of two\nstages: Stage 1 adapts models to cross-domain topological discrepancies using\nthe proposed Topological Meta Difference Convolutions (TopoMDCs), which enhance\ntopological representation without altering pre-trained parameters; Stage 2\nimproves topological continuity by a novel Topology Hard sample Generation\n(TopoHG) strategy and prediction alignment on hard samples with pseudo-labels\nin the generated pseudo-break regions. Extensive experiments across four\nscenarios and ten datasets demonstrate TopoTTA's effectiveness in handling\ntopological distribution shifts, achieving an average improvement of 31.81% in\nclDice. TopoTTA also serves as a plug-and-play TTA solution for CNN-based TSS\nmodels.", 'authors': ['Jiale Zhou', 'Wenhan Wang', 'Shikun Li', 'Xiaolei Qu', 'Xin Guo', 'Yizhong Liu', 'Wenzhong Tang', 'Xun Lin', 'Yefeng Zheng'], 'published': '2025-08-01 08:59:13+00:00', 'categories': ['cs.CV', 'cs.AI'], 'url': 'http://arxiv.org/abs/2508.00442v1'}
259
2508.00298
2,508.00298
null
AniMer+: Unified Pose and Shape Estimation Across Mammalia and Aves via Family-Aware Transformer
In the era of foundation models, achieving a unified understanding of different dynamic objects through a single network has the potential to empower stronger spatial intelligence. Moreover, accurate estimation of animal pose and shape across diverse species is essential for quantitative analysis in biological research. However, this topic remains underexplored due to the limited network capacity of previous methods and the scarcity of comprehensive multi-species datasets. To address these limitations, we introduce AniMer+, an extended version of our scalable AniMer framework. In this paper, we focus on a unified approach for reconstructing mammals (mammalia) and birds (aves). A key innovation of AniMer+ is its high-capacity, family-aware Vision Transformer (ViT) incorporating a Mixture-of-Experts (MoE) design. Its architecture partitions network layers into taxa-specific components (for mammalia and aves) and taxa-shared components, enabling efficient learning of both distinct and common anatomical features within a single model. To overcome the critical shortage of 3D training data, especially for birds, we introduce a diffusion-based conditional image generation pipeline. This pipeline produces two large-scale synthetic datasets: CtrlAni3D for quadrupeds and CtrlAVES3D for birds. To note, CtrlAVES3D is the first large-scale, 3D-annotated dataset for birds, which is crucial for resolving single-view depth ambiguities. Trained on an aggregated collection of 41.3k mammalian and 12.4k avian images (combining real and synthetic data), our method demonstrates superior performance over existing approaches across a wide range of benchmarks, including the challenging out-of-domain Animal Kingdom dataset. Ablation studies confirm the effectiveness of both our novel network architecture and the generated synthetic datasets in enhancing real-world application performance.
2025-08-01
{'id': '2508.00298', 'title': 'AniMer+: Unified Pose and Shape Estimation Across Mammalia and Aves via Family-Aware Transformer', 'abstract': 'In the era of foundation models, achieving a unified understanding of\ndifferent dynamic objects through a single network has the potential to empower\nstronger spatial intelligence. Moreover, accurate estimation of animal pose and\nshape across diverse species is essential for quantitative analysis in\nbiological research. However, this topic remains underexplored due to the\nlimited network capacity of previous methods and the scarcity of comprehensive\nmulti-species datasets. To address these limitations, we introduce AniMer+, an\nextended version of our scalable AniMer framework. In this paper, we focus on a\nunified approach for reconstructing mammals (mammalia) and birds (aves). A key\ninnovation of AniMer+ is its high-capacity, family-aware Vision Transformer\n(ViT) incorporating a Mixture-of-Experts (MoE) design. Its architecture\npartitions network layers into taxa-specific components (for mammalia and aves)\nand taxa-shared components, enabling efficient learning of both distinct and\ncommon anatomical features within a single model. To overcome the critical\nshortage of 3D training data, especially for birds, we introduce a\ndiffusion-based conditional image generation pipeline. This pipeline produces\ntwo large-scale synthetic datasets: CtrlAni3D for quadrupeds and CtrlAVES3D for\nbirds. To note, CtrlAVES3D is the first large-scale, 3D-annotated dataset for\nbirds, which is crucial for resolving single-view depth ambiguities. Trained on\nan aggregated collection of 41.3k mammalian and 12.4k avian images (combining\nreal and synthetic data), our method demonstrates superior performance over\nexisting approaches across a wide range of benchmarks, including the\nchallenging out-of-domain Animal Kingdom dataset. Ablation studies confirm the\neffectiveness of both our novel network architecture and the generated\nsynthetic datasets in enhancing real-world application performance.', 'authors': ['Jin Lyu', 'Liang An', 'Li Lin', 'Pujin Cheng', 'Yebin Liu', 'Xiaoying Tang'], 'published': '2025-08-01 03:53:03+00:00', 'categories': ['cs.CV'], 'url': 'http://arxiv.org/abs/2508.00298v1'}
260
2508.00462
2,508.00462
null
Managing Power Gaps as a Topic of Pair Programming Skill: A Grounded Theory
Context: Pair Programming as a work mode is used (occasionally or frequently) throughout professional software development. Objective: Understand what power-related phenomena occur in pair programming as it is used in industry; give advice to practitioners on how to do better pair programming. Method: Analyze 22 industrial pair programming sessions using Grounded Theory Methodology. Formulate a Grounded Theory on power-related behaviors. Run a survey with 292 participants about that theory. Use it to demonstrate that the phenomena are common. Results: Our theory describes the phenomenon of Power Gap: a perceived difference in participation opportunities. The theory shows the behaviors that create a Power Gap or result from it. Power Gaps tend to damage knowledge transfer, code quality, and process effi ciency. The survey results show that all concepts from our theory are frequent in practice. They also provide more grounding for concepts that are observable only indirectly. Conclusions: It is a valuable component of pair programming skill to be able to avoid Power Gaps. Specifically, pair partners need to avoid Hierarchical Behavior (which tends to create or increase a Power Gap) and should perform enough Equalizing Behavior (which prevents or reduces a Power Gap).
2025-08-01
{'id': '2508.00462', 'title': 'Managing Power Gaps as a Topic of Pair Programming Skill: A Grounded Theory', 'abstract': 'Context: Pair Programming as a work mode is used (occasionally or frequently)\nthroughout professional software development. Objective: Understand what\npower-related phenomena occur in pair programming as it is used in industry;\ngive advice to practitioners on how to do better pair programming. Method:\nAnalyze 22 industrial pair programming sessions using Grounded Theory\nMethodology. Formulate a Grounded Theory on power-related behaviors. Run a\nsurvey with 292 participants about that theory. Use it to demonstrate that the\nphenomena are common. Results: Our theory describes the phenomenon of Power\nGap: a perceived difference in participation opportunities. The theory shows\nthe behaviors that create a Power Gap or result from it. Power Gaps tend to\ndamage knowledge transfer, code quality, and process effi ciency. The survey\nresults show that all concepts from our theory are frequent in practice. They\nalso provide more grounding for concepts that are observable only indirectly.\nConclusions: It is a valuable component of pair programming skill to be able to\navoid Power Gaps. Specifically, pair partners need to avoid Hierarchical\nBehavior (which tends to create or increase a Power Gap) and should perform\nenough Equalizing Behavior (which prevents or reduces a Power Gap).', 'authors': ['Linus Ververs', 'Lutz Prechelt'], 'published': '2025-08-01 09:34:20+00:00', 'categories': ['cs.SE'], 'url': 'http://arxiv.org/abs/2508.00462v1'}
261
2508.00512
2,508.00512
null
Projective Delineability for Single Cell Construction
The cylindrical algebraic decomposition (CAD) is the only complete method used in practice for solving problems like quantifier elimination or SMT solving related to real algebra, despite its doubly exponential complexity. Recent exploration-guided algorithms like NLSAT, NuCAD, and CAlC rely on CAD technology but reduce the computational effort heuristically. Single cell construction is a paradigm that is used in each of these algorithms. The central property on which the CAD algorithm is based is called delineability. Recently, we introduced a weaker notion called projective delineability which can require fewer computations to guarantee, but needs to be applied carefully. This paper adapts the single cell construction for exploiting projective delineability and reports on experimental results.
2025-08-01
{'id': '2508.00512', 'title': 'Projective Delineability for Single Cell Construction', 'abstract': 'The cylindrical algebraic decomposition (CAD) is the only complete method\nused in practice for solving problems like quantifier elimination or SMT\nsolving related to real algebra, despite its doubly exponential complexity.\nRecent exploration-guided algorithms like NLSAT, NuCAD, and CAlC rely on CAD\ntechnology but reduce the computational effort heuristically. Single cell\nconstruction is a paradigm that is used in each of these algorithms.\n The central property on which the CAD algorithm is based is called\ndelineability. Recently, we introduced a weaker notion called projective\ndelineability which can require fewer computations to guarantee, but needs to\nbe applied carefully. This paper adapts the single cell construction for\nexploiting projective delineability and reports on experimental results.', 'authors': ['Jasper Nalbach', 'Lucas Michel', 'Erika Ábrahám', 'Christopher W. Brown', 'James H. Davenport', 'Matthew England', 'Pierre Mathonet', 'Naïm Zénaïdi'], 'published': '2025-08-01 10:45:24+00:00', 'categories': ['cs.SC'], 'url': 'http://arxiv.org/abs/2508.00512v1'}
262
2508.00729
2,508.00729
null
Graphs with tree decompositions of small graphs and realizing them as the Reeb graphs of real algebraic functions
We have been interested in graphs and realizing them as Reeb graphs of explicit real algebraic functions. The Reeb graph of a differentiable function is the quotient space of the manifold of the domain, regarded as the space consisting of all components of preimages of all single points. Reeb graphs have been fundamental and strong tools in geometry of manifolds since the birth of theory of Morse functions, in the former half of the 20th century. We can easily see that the Reeb graph of the natural height of the unit sphere whose dimension is at least $2$ is a graph with exactly one edge and two edges. We are concerned with realizations of graphs decomposed into trees nicely, each vertex of which corresponds to a graph with exactly one edge and two edges or a graph with exactly two edges homeomorphic to a circle.
2025-08-01
{'id': '2508.00729', 'title': 'Graphs with tree decompositions of small graphs and realizing them as the Reeb graphs of real algebraic functions', 'abstract': 'We have been interested in graphs and realizing them as Reeb graphs of\nexplicit real algebraic functions.\n The Reeb graph of a differentiable function is the quotient space of the\nmanifold of the domain, regarded as the space consisting of all components of\npreimages of all single points. Reeb graphs have been fundamental and strong\ntools in geometry of manifolds since the birth of theory of Morse functions, in\nthe former half of the 20th century. We can easily see that the Reeb graph of\nthe natural height of the unit sphere whose dimension is at least $2$ is a\ngraph with exactly one edge and two edges.\n We are concerned with realizations of graphs decomposed into trees nicely,\neach vertex of which corresponds to a graph with exactly one edge and two edges\nor a graph with exactly two edges homeomorphic to a circle.', 'authors': ['Naoki Kitazawa'], 'published': '2025-08-01 15:51:54+00:00', 'categories': ['math.AG', 'math.CO', 'math.MG'], 'url': 'http://arxiv.org/abs/2508.00729v2'}
264
2508.00762
2,508.00762
null
ITUNLP at SemEval-2025 Task 8: Question-Answering over Tabular Data: A Zero-Shot Approach using LLM-Driven Code Generation
This paper presents our system for SemEval-2025 Task 8: DataBench, Question-Answering over Tabular Data. The primary objective of this task is to perform question answering on given tabular datasets from diverse domains under two subtasks: DataBench QA (Subtask I) and DataBench Lite QA (Subtask II). To tackle both subtasks, we developed a zero-shot solution with a particular emphasis on leveraging Large Language Model (LLM)-based code generation. Specifically, we propose a Python code generation framework utilizing state-of-the-art open-source LLMs to generate executable Pandas code via optimized prompting strategies. Our experiments reveal that different LLMs exhibit varying levels of effectiveness in Python code generation. Additionally, results show that Python code generation achieves superior performance in tabular question answering compared to alternative approaches. Although our ranking among zero-shot systems is unknown at the time of this paper's submission, our system achieved eighth place in Subtask I and sixth place in Subtask~II among the 30 systems that outperformed the baseline in the open-source models category.
2025-08-01
{'id': '2508.00762', 'title': 'ITUNLP at SemEval-2025 Task 8: Question-Answering over Tabular Data: A Zero-Shot Approach using LLM-Driven Code Generation', 'abstract': "This paper presents our system for SemEval-2025 Task 8: DataBench,\nQuestion-Answering over Tabular Data. The primary objective of this task is to\nperform question answering on given tabular datasets from diverse domains under\ntwo subtasks: DataBench QA (Subtask I) and DataBench Lite QA (Subtask II). To\ntackle both subtasks, we developed a zero-shot solution with a particular\nemphasis on leveraging Large Language Model (LLM)-based code generation.\nSpecifically, we propose a Python code generation framework utilizing\nstate-of-the-art open-source LLMs to generate executable Pandas code via\noptimized prompting strategies. Our experiments reveal that different LLMs\nexhibit varying levels of effectiveness in Python code generation.\nAdditionally, results show that Python code generation achieves superior\nperformance in tabular question answering compared to alternative approaches.\nAlthough our ranking among zero-shot systems is unknown at the time of this\npaper's submission, our system achieved eighth place in Subtask I and sixth\nplace in Subtask~II among the 30 systems that outperformed the baseline in the\nopen-source models category.", 'authors': ['Atakan Site', 'Emre Hakan Erdemir', 'Gülşen Eryiğit'], 'published': '2025-08-01 16:38:18+00:00', 'categories': ['cs.CL'], 'url': 'http://arxiv.org/abs/2508.00762v1'}
265
2508.00717
2,508.00717
null
Generative AI in Higher Education: Evidence from an Elite College
Generative AI is transforming higher education, yet systematic evidence on student adoption remains limited. Using novel survey data from a selective U.S. college, we document over 80 percent of students using AI academically within two years of ChatGPT's release. Adoption varies across disciplines, demographics, and achievement levels, highlighting AI's potential to reshape educational inequalities. Students predominantly use AI for augmenting learning (e.g., explanations, feedback), but also to automate tasks (e.g., essay generation). Positive perceptions of AI's educational benefits strongly predict adoption. Institutional policies can influence usage patterns but risk creating unintended disparate impacts across student groups due to uneven compliance.
2025-08-01
{'id': '2508.00717', 'title': 'Generative AI in Higher Education: Evidence from an Elite College', 'abstract': "Generative AI is transforming higher education, yet systematic evidence on\nstudent adoption remains limited. Using novel survey data from a selective U.S.\ncollege, we document over 80 percent of students using AI academically within\ntwo years of ChatGPT's release. Adoption varies across disciplines,\ndemographics, and achievement levels, highlighting AI's potential to reshape\neducational inequalities. Students predominantly use AI for augmenting learning\n(e.g., explanations, feedback), but also to automate tasks (e.g., essay\ngeneration). Positive perceptions of AI's educational benefits strongly predict\nadoption. Institutional policies can influence usage patterns but risk creating\nunintended disparate impacts across student groups due to uneven compliance.", 'authors': ['Zara Contractor', 'Germán Reyes'], 'published': '2025-08-01 15:35:03+00:00', 'categories': ['econ.GN', 'q-fin.EC'], 'url': 'http://arxiv.org/abs/2508.00717v1'}
266
2508.00534
2,508.00534
null
Towards a unified framework for programming paradigms: A systematic review of classification formalisms and methodological foundations
The rise of multi-paradigm languages challenges traditional classification methods, leading to practical software engineering issues like interoperability defects. This systematic literature review (SLR) maps the formal foundations of programming paradigms. Our objective is twofold: (1) to assess the state of the art of classification formalisms and their limitations, and (2) to identify the conceptual primitives and mathematical frameworks for a more powerful, reconstructive approach. Based on a synthesis of 74 primary studies, we find that existing taxonomies lack conceptual granularity, a unified formal basis, and struggle with hybrid languages. In response, our analysis reveals a strong convergence toward a compositional reconstruction of paradigms. This approach identifies a minimal set of orthogonal, atomic primitives and leverages mathematical frameworks, predominantly Type theory, Category theory and Unifying Theories of Programming (UTP), to formally guarantee their compositional properties. We conclude that the literature reflects a significant intellectual shift away from classification towards these promising formal, reconstructive frameworks. This review provides a map of this evolution and proposes a research agenda for their unification.
2025-08-01
{'id': '2508.00534', 'title': 'Towards a unified framework for programming paradigms: A systematic review of classification formalisms and methodological foundations', 'abstract': 'The rise of multi-paradigm languages challenges traditional classification\nmethods, leading to practical software engineering issues like interoperability\ndefects. This systematic literature review (SLR) maps the formal foundations of\nprogramming paradigms. Our objective is twofold: (1) to assess the state of the\nart of classification formalisms and their limitations, and (2) to identify the\nconceptual primitives and mathematical frameworks for a more powerful,\nreconstructive approach.\n Based on a synthesis of 74 primary studies, we find that existing taxonomies\nlack conceptual granularity, a unified formal basis, and struggle with hybrid\nlanguages. In response, our analysis reveals a strong convergence toward a\ncompositional reconstruction of paradigms. This approach identifies a minimal\nset of orthogonal, atomic primitives and leverages mathematical frameworks,\npredominantly Type theory, Category theory and Unifying Theories of Programming\n(UTP), to formally guarantee their compositional properties.\n We conclude that the literature reflects a significant intellectual shift\naway from classification towards these promising formal, reconstructive\nframeworks. This review provides a map of this evolution and proposes a\nresearch agenda for their unification.', 'authors': ['Mikel Vandeloise'], 'published': '2025-08-01 11:19:40+00:00', 'categories': ['cs.PL', 'cs.CL', 'D.3.2; F.3.2; D.3.1'], 'url': 'http://arxiv.org/abs/2508.00534v1'}
267
2508.00467
2,508.00467
null
SubCDM: Collective Decision-Making with a Swarm Subset
Collective decision-making is a key function of autonomous robot swarms, enabling them to reach a consensus on actions based on environmental features. Existing strategies require the participation of all robots in the decision-making process, which is resource-intensive and prevents the swarm from allocating the robots to any other tasks. We propose Subset-Based Collective Decision-Making (SubCDM), which enables decisions using only a swarm subset. The construction of the subset is dynamic and decentralized, relying solely on local information. Our method allows the swarm to adaptively determine the size of the subset for accurate decision-making, depending on the difficulty of reaching a consensus. Simulation results using one hundred robots show that our approach achieves accuracy comparable to using the entire swarm while reducing the number of robots required to perform collective decision-making, making it a resource-efficient solution for collective decision-making in swarm robotics.
2025-08-01
{'id': '2508.00467', 'title': 'SubCDM: Collective Decision-Making with a Swarm Subset', 'abstract': 'Collective decision-making is a key function of autonomous robot swarms,\nenabling them to reach a consensus on actions based on environmental features.\nExisting strategies require the participation of all robots in the\ndecision-making process, which is resource-intensive and prevents the swarm\nfrom allocating the robots to any other tasks. We propose Subset-Based\nCollective Decision-Making (SubCDM), which enables decisions using only a swarm\nsubset. The construction of the subset is dynamic and decentralized, relying\nsolely on local information. Our method allows the swarm to adaptively\ndetermine the size of the subset for accurate decision-making, depending on the\ndifficulty of reaching a consensus. Simulation results using one hundred robots\nshow that our approach achieves accuracy comparable to using the entire swarm\nwhile reducing the number of robots required to perform collective\ndecision-making, making it a resource-efficient solution for collective\ndecision-making in swarm robotics.', 'authors': ['Samratul Fuady', 'Danesh Tarapore', 'Mohammad D. Soorati'], 'published': '2025-08-01 09:40:38+00:00', 'categories': ['cs.RO'], 'url': 'http://arxiv.org/abs/2508.00467v1'}
273
2508.05634v1
2,508.05634
1
Towards Generalizable Safety in Crowd Navigation via Conformal Uncertainty Handling
Mobile robots navigating in crowds trained using reinforcement learning are known to suffer performance degradation when faced with out-of-distribution scenarios. We propose that by properly accounting for the uncertainties of pedestrians, a robot can learn safe navigation policies that are robust to distribution shifts. Our method augments agent observations with prediction uncertainty estimates generated by adaptive conformal inference, and it uses these estimates to guide the agent's behavior through constrained reinforcement learning. The system helps regulate the agent's actions and enables it to adapt to distribution shifts. In the in-distribution setting, our approach achieves a 96.93% success rate, which is over 8.80% higher than the previous state-of-the-art baselines with over 3.72 times fewer collisions and 2.43 times fewer intrusions into ground-truth human future trajectories. In three out-of-distribution scenarios, our method shows much stronger robustness when facing distribution shifts in velocity variations, policy changes, and transitions from individual to group dynamics. We deploy our method on a real robot, and experiments show that the robot makes safe and robust decisions when interacting with both sparse and dense crowds. Our code and videos are available on https://gen-safe-nav.github.io/.
2025-08-07
{'id': '2508.05634v1', 'title': 'Towards Generalizable Safety in Crowd Navigation via Conformal Uncertainty Handling', 'abstract': "Mobile robots navigating in crowds trained using reinforcement learning are\nknown to suffer performance degradation when faced with out-of-distribution\nscenarios. We propose that by properly accounting for the uncertainties of\npedestrians, a robot can learn safe navigation policies that are robust to\ndistribution shifts. Our method augments agent observations with prediction\nuncertainty estimates generated by adaptive conformal inference, and it uses\nthese estimates to guide the agent's behavior through constrained reinforcement\nlearning. The system helps regulate the agent's actions and enables it to adapt\nto distribution shifts. In the in-distribution setting, our approach achieves a\n96.93% success rate, which is over 8.80% higher than the previous\nstate-of-the-art baselines with over 3.72 times fewer collisions and 2.43 times\nfewer intrusions into ground-truth human future trajectories. In three\nout-of-distribution scenarios, our method shows much stronger robustness when\nfacing distribution shifts in velocity variations, policy changes, and\ntransitions from individual to group dynamics. We deploy our method on a real\nrobot, and experiments show that the robot makes safe and robust decisions when\ninteracting with both sparse and dense crowds. Our code and videos are\navailable on https://gen-safe-nav.github.io/.", 'authors': ['Jianpeng Yao', 'Xiaopan Zhang', 'Yu Xia', 'Zejin Wang', 'Amit K. Roy-Chowdhury', 'Jiachen Li'], 'published': '2025-08-07 17:59:43+00:00', 'categories': ['cs.RO', 'cs.AI', 'cs.CV', 'cs.LG', 'cs.SY', 'eess.SY'], 'url': 'http://arxiv.org/abs/2508.05634v1'}
268
2508.00688
2,508.00688
null
Criticality-Based Dynamic Topology Optimization for Enhancing Aerial-Marine Swarm Resilience
Heterogeneous marine-aerial swarm networks encounter substantial difficulties due to targeted communication disruptions and structural weaknesses in adversarial environments. This paper proposes a two-step framework to strengthen the network's resilience. Specifically, our framework combines the node prioritization based on criticality with multi-objective topology optimization. First, we design a three-layer architecture to represent structural, communication, and task dependencies of the swarm networks. Then, we introduce the SurBi-Ranking method, which utilizes graph convolutional networks, to dynamically evaluate and rank the criticality of nodes and edges in real time. Next, we apply the NSGA-III algorithm to optimize the network topology, aiming to balance communication efficiency, global connectivity, and mission success rate. Experiments demonstrate that compared to traditional methods like K-Shell, our SurBi-Ranking method identifies critical nodes and edges with greater accuracy, as deliberate attacks on these components cause more significant connectivity degradation. Furthermore, our optimization approach, when prioritizing SurBi-Ranked critical components under attack, reduces the natural connectivity degradation by around 30%, achieves higher mission success rates, and incurs lower communication reconfiguration costs, ensuring sustained connectivity and mission effectiveness across multi-phase operations.
2025-08-01
{'id': '2508.00688', 'title': 'Criticality-Based Dynamic Topology Optimization for Enhancing Aerial-Marine Swarm Resilience', 'abstract': "Heterogeneous marine-aerial swarm networks encounter substantial difficulties\ndue to targeted communication disruptions and structural weaknesses in\nadversarial environments. This paper proposes a two-step framework to\nstrengthen the network's resilience. Specifically, our framework combines the\nnode prioritization based on criticality with multi-objective topology\noptimization. First, we design a three-layer architecture to represent\nstructural, communication, and task dependencies of the swarm networks. Then,\nwe introduce the SurBi-Ranking method, which utilizes graph convolutional\nnetworks, to dynamically evaluate and rank the criticality of nodes and edges\nin real time. Next, we apply the NSGA-III algorithm to optimize the network\ntopology, aiming to balance communication efficiency, global connectivity, and\nmission success rate. Experiments demonstrate that compared to traditional\nmethods like K-Shell, our SurBi-Ranking method identifies critical nodes and\nedges with greater accuracy, as deliberate attacks on these components cause\nmore significant connectivity degradation. Furthermore, our optimization\napproach, when prioritizing SurBi-Ranked critical components under attack,\nreduces the natural connectivity degradation by around 30%, achieves higher\nmission success rates, and incurs lower communication reconfiguration costs,\nensuring sustained connectivity and mission effectiveness across multi-phase\noperations.", 'authors': ['Ruiyang Huang', 'Haocheng Wang', 'Yixuan Shen', 'Ning Gao', 'Qiang Ni', 'Shi Jin', 'Yifan Wu'], 'published': '2025-08-01 15:03:23+00:00', 'categories': ['cs.NI', 'eess.SP'], 'url': 'http://arxiv.org/abs/2508.00688v1'}
269
2508.00751
2,508.00751
null
Harnessing the Power of Interleaving and Counterfactual Evaluation for Airbnb Search Ranking
Evaluation plays a crucial role in the development of ranking algorithms on search and recommender systems. It enables online platforms to create user-friendly features that drive commercial success in a steady and effective manner. The online environment is particularly conducive to applying causal inference techniques, such as randomized controlled experiments (known as A/B test), which are often more challenging to implement in fields like medicine and public policy. However, businesses face unique challenges when it comes to effective A/B test. Specifically, achieving sufficient statistical power for conversion-based metrics can be time-consuming, especially for significant purchases like booking accommodations. While offline evaluations are quicker and more cost-effective, they often lack accuracy and are inadequate for selecting candidates for A/B test. To address these challenges, we developed interleaving and counterfactual evaluation methods to facilitate rapid online assessments for identifying the most promising candidates for A/B tests. Our approach not only increased the sensitivity of experiments by a factor of up to 100 (depending on the approach and metrics) compared to traditional A/B testing but also streamlined the experimental process. The practical insights gained from usage in production can also benefit organizations with similar interests.
2025-08-01
{'id': '2508.00751', 'title': 'Harnessing the Power of Interleaving and Counterfactual Evaluation for Airbnb Search Ranking', 'abstract': 'Evaluation plays a crucial role in the development of ranking algorithms on\nsearch and recommender systems. It enables online platforms to create\nuser-friendly features that drive commercial success in a steady and effective\nmanner. The online environment is particularly conducive to applying causal\ninference techniques, such as randomized controlled experiments (known as A/B\ntest), which are often more challenging to implement in fields like medicine\nand public policy. However, businesses face unique challenges when it comes to\neffective A/B test. Specifically, achieving sufficient statistical power for\nconversion-based metrics can be time-consuming, especially for significant\npurchases like booking accommodations. While offline evaluations are quicker\nand more cost-effective, they often lack accuracy and are inadequate for\nselecting candidates for A/B test. To address these challenges, we developed\ninterleaving and counterfactual evaluation methods to facilitate rapid online\nassessments for identifying the most promising candidates for A/B tests. Our\napproach not only increased the sensitivity of experiments by a factor of up to\n100 (depending on the approach and metrics) compared to traditional A/B testing\nbut also streamlined the experimental process. The practical insights gained\nfrom usage in production can also benefit organizations with similar interests.', 'authors': ['Qing Zhang', 'Alex Deng', 'Michelle Du', 'Huiji Gao', 'Liwei He', 'Sanjeev Katariya'], 'published': '2025-08-01 16:28:18+00:00', 'categories': ['cs.IR', 'cs.AI', 'H.3; G.3'], 'url': 'http://arxiv.org/abs/2508.00751v1'}
270
2508.00309
2,508.00309
null
Helicity Fluxes and Hemispheric Helicity Rule of Active Regions Emerging from the Convection Zone Dynamo
Using a 3D non-linear mean-field solar dynamo model, we investigate the magnetic helicity flux and magnetic twist, and tilt parameters of bipolar magnetic regions (BMRs) emerging from the solar convection zone due to the magnetic buoyancy instability. The twist and tilt of the BMR magnetic field are modeled as a result of an effective electromotive force along the rising part of the toroidal magnetic field. This force generates the poloidal field that tilts the whole magnetic configuration. We find that variations of BMR's twist and tilt determine the magnitude and the sign of the magnetic helicity flux on the solar surface. The model shows that the helicity flux associated with the BMR's tilt/twist is the dominant contribution to the BMR helicity at the beginning of the BMR's evolution, while the effect of differential rotation is the main source of the helicity flux at the final stage of the BMR's evolution. We discuss the implications of these effects on the basic properties and variations of the hemispheric helicity rule of active regions on the solar surface.
2025-08-01
{'id': '2508.00309', 'title': 'Helicity Fluxes and Hemispheric Helicity Rule of Active Regions Emerging from the Convection Zone Dynamo', 'abstract': "Using a 3D non-linear mean-field solar dynamo model, we investigate the\nmagnetic helicity flux and magnetic twist, and tilt parameters of bipolar\nmagnetic regions (BMRs) emerging from the solar convection zone due to the\nmagnetic buoyancy instability. The twist and tilt of the BMR magnetic field are\nmodeled as a result of an effective electromotive force along the rising part\nof the toroidal magnetic field. This force generates the poloidal field that\ntilts the whole magnetic configuration. We find that variations of BMR's twist\nand tilt determine the magnitude and the sign of the magnetic helicity flux on\nthe solar surface. The model shows that the helicity flux associated with the\nBMR's tilt/twist is the dominant contribution to the BMR helicity at the\nbeginning of the BMR's evolution, while the effect of differential rotation is\nthe main source of the helicity flux at the final stage of the BMR's evolution.\nWe discuss the implications of these effects on the basic properties and\nvariations of the hemispheric helicity rule of active regions on the solar\nsurface.", 'authors': ['V. V. Pipin', 'Shangbin Yang', 'A. G. Kosovichev'], 'published': '2025-08-01 04:25:55+00:00', 'categories': ['astro-ph.SR'], 'url': 'http://arxiv.org/abs/2508.00309v1'}
271
2508.00421
2,508.00421
null
UIS-Mamba: Exploring Mamba for Underwater Instance Segmentation via Dynamic Tree Scan and Hidden State Weaken
Underwater Instance Segmentation (UIS) tasks are crucial for underwater complex scene detection. Mamba, as an emerging state space model with inherently linear complexity and global receptive fields, is highly suitable for processing image segmentation tasks with long sequence features. However, due to the particularity of underwater scenes, there are many challenges in applying Mamba to UIS. The existing fixed-patch scanning mechanism cannot maintain the internal continuity of scanned instances in the presence of severely underwater color distortion and blurred instance boundaries, and the hidden state of the complex underwater background can also inhibit the understanding of instance objects. In this work, we propose the first Mamba-based underwater instance segmentation model UIS-Mamba, and design two innovative modules, Dynamic Tree Scan (DTS) and Hidden State Weaken (HSW), to migrate Mamba to the underwater task. DTS module maintains the continuity of the internal features of the instance objects by allowing the patches to dynamically offset and scale, thereby guiding the minimum spanning tree and providing dynamic local receptive fields. HSW module suppresses the interference of complex backgrounds and effectively focuses the information flow of state propagation to the instances themselves through the Ncut-based hidden state weakening mechanism. Experimental results show that UIS-Mamba achieves state-of-the-art performance on both UIIS and USIS10K datasets, while maintaining a low number of parameters and computational complexity. Code is available at https://github.com/Maricalce/UIS-Mamba.
2025-08-01
{'id': '2508.00421', 'title': 'UIS-Mamba: Exploring Mamba for Underwater Instance Segmentation via Dynamic Tree Scan and Hidden State Weaken', 'abstract': 'Underwater Instance Segmentation (UIS) tasks are crucial for underwater\ncomplex scene detection. Mamba, as an emerging state space model with\ninherently linear complexity and global receptive fields, is highly suitable\nfor processing image segmentation tasks with long sequence features. However,\ndue to the particularity of underwater scenes, there are many challenges in\napplying Mamba to UIS. The existing fixed-patch scanning mechanism cannot\nmaintain the internal continuity of scanned instances in the presence of\nseverely underwater color distortion and blurred instance boundaries, and the\nhidden state of the complex underwater background can also inhibit the\nunderstanding of instance objects. In this work, we propose the first\nMamba-based underwater instance segmentation model UIS-Mamba, and design two\ninnovative modules, Dynamic Tree Scan (DTS) and Hidden State Weaken (HSW), to\nmigrate Mamba to the underwater task. DTS module maintains the continuity of\nthe internal features of the instance objects by allowing the patches to\ndynamically offset and scale, thereby guiding the minimum spanning tree and\nproviding dynamic local receptive fields. HSW module suppresses the\ninterference of complex backgrounds and effectively focuses the information\nflow of state propagation to the instances themselves through the Ncut-based\nhidden state weakening mechanism. Experimental results show that UIS-Mamba\nachieves state-of-the-art performance on both UIIS and USIS10K datasets, while\nmaintaining a low number of parameters and computational complexity. Code is\navailable at https://github.com/Maricalce/UIS-Mamba.', 'authors': ['Runmin Cong', 'Zongji Yu', 'Hao Fang', 'Haoyan Sun', 'Sam Kwong'], 'published': '2025-08-01 08:21:24+00:00', 'categories': ['cs.CV'], 'url': 'http://arxiv.org/abs/2508.00421v1'}
272
2508.00503
2,508.00503
null
Using tunable coherence for reaching micrometer coherence lengths and suppressing stray light in a power-recycled Michelson interferometer
By reentering into laser interferometers, scattered or stray light introduces non-linear noise. This is a major limitation of precision interferometers as preventing such parasitic light is nearly impossible. Thus, substantial effort is put into mitigating the reentering of these fields in various ways. Ground-based laser interferometric gravitational wave detectors employ such mitigation techniques to reduce otherwise restrictive stray light noise. However, they are now reaching sensitivities where conventional mitigation techniques reach limitations. Further improvements planed for future observatories are placing even more demanding constraints on tolerable stray light power. We previously presented tunable coherence as a possible technique to ease these constraints and suppress unwanted coherent interference. For these promising demonstrations, the remaining coherence length and achievable suppression in length-constrained layouts was limited, among other things, by the used pseudo-random-noise phase modulation frequency. In this work, we demonstrate stray light suppression and cavity performance at modulation frequencies up to 10 GHz. This reduces the remaining coherence to a few centimeter in an interferometer, and even to the scale of the laser wavelength in a cavity. We further present a first demonstration of tunable coherence in a power-recycled Michelson interferometer, successfully suppressing stray light in a more complex topology.
2025-08-01
{'id': '2508.00503', 'title': 'Using tunable coherence for reaching micrometer coherence lengths and suppressing stray light in a power-recycled Michelson interferometer', 'abstract': 'By reentering into laser interferometers, scattered or stray light introduces\nnon-linear noise. This is a major limitation of precision interferometers as\npreventing such parasitic light is nearly impossible. Thus, substantial effort\nis put into mitigating the reentering of these fields in various ways.\nGround-based laser interferometric gravitational wave detectors employ such\nmitigation techniques to reduce otherwise restrictive stray light noise.\nHowever, they are now reaching sensitivities where conventional mitigation\ntechniques reach limitations. Further improvements planed for future\nobservatories are placing even more demanding constraints on tolerable stray\nlight power. We previously presented tunable coherence as a possible technique\nto ease these constraints and suppress unwanted coherent interference. For\nthese promising demonstrations, the remaining coherence length and achievable\nsuppression in length-constrained layouts was limited, among other things, by\nthe used pseudo-random-noise phase modulation frequency. In this work, we\ndemonstrate stray light suppression and cavity performance at modulation\nfrequencies up to 10 GHz. This reduces the remaining coherence to a few\ncentimeter in an interferometer, and even to the scale of the laser wavelength\nin a cavity. We further present a first demonstration of tunable coherence in a\npower-recycled Michelson interferometer, successfully suppressing stray light\nin a more complex topology.', 'authors': ['Daniel Voigt', 'Oliver Gerberding'], 'published': '2025-08-01 10:31:09+00:00', 'categories': ['physics.optics', 'astro-ph.IM', 'gr-qc', 'physics.ins-det'], 'url': 'http://arxiv.org/abs/2508.00503v1'}
274
2508.00389
2,508.00389
null
Discrete frames of non-uniform shifts in frequency domain
Frames in separable Hilbert spaces gives stable analysis and reconstruction of each vector in the underlying space. In this paper, we study frame conditions for a collection of matrix-valued functions obtained by non-uniform shifts. We give necessary and sufficient conditions for the existence of matrix-valued discrete Bessel sequence over non-uniform displacement parameters in terms of the Fourier transform of window functions. We also present perturbation results for matrix-valued non-uniform discrete frames.
2025-08-01
{'id': '2508.00389', 'title': 'Discrete frames of non-uniform shifts in frequency domain', 'abstract': 'Frames in separable Hilbert spaces gives stable analysis and reconstruction\nof each vector in the underlying space. In this paper, we study frame\nconditions for a collection of matrix-valued functions obtained by non-uniform\nshifts. We give necessary and sufficient conditions for the existence of\nmatrix-valued discrete Bessel sequence over non-uniform displacement parameters\nin terms of the Fourier transform of window functions. We also present\nperturbation results for matrix-valued non-uniform discrete frames.', 'authors': ['Hari Krishan Malhotra', 'Manisha Chhillar', 'Lalit Kumar Vashisht'], 'published': '2025-08-01 07:34:21+00:00', 'categories': ['math.FA'], 'url': 'http://arxiv.org/abs/2508.00389v1'}
275
2508.00335
2,508.00335
null
Proper Orthogonal Decomposition-based Model-Order Reduction for Smoothed Particle Hydrodynamics Simulation -- Mass-Spring-Damper System
Model Order Reduction (MOR) based on Proper Orthogonal Decomposition (POD) and Smooth Particle Hydrodynamics (SPH) has proven effective in various applications. Most MOR methods utilizing POD are implemented within a pure Eulerian framework, while significantly less attention has been given to POD in a Lagrangian context. In this paper, we present the POD-MOR of SPH simulations applied to a mass-spring-damper system with two primary objectives: 1. To evaluate the performance of the data-driven POD-MOR approach. 2. To investigate potential methods for accelerating POD-MOR computations. Although the mass-spring-damper system is linear, its SPH implementations are nonlinear, and POD-MOR does not automatically lead to faster computations. Our findings indicate that (1) the POD-MOR effectively reduces the degrees of freedom in the SPH simulations by capturing the essential modes, and (2) in various cases, the acceleration of POD-MOR can be achieved without compromising accuracy. We hope that our results will motivate further investigations into the design of POD-MOR algorithms for nonlinear Lagrangian systems.
2025-08-01
{'id': '2508.00335', 'title': 'Proper Orthogonal Decomposition-based Model-Order Reduction for Smoothed Particle Hydrodynamics Simulation -- Mass-Spring-Damper System', 'abstract': 'Model Order Reduction (MOR) based on Proper Orthogonal Decomposition (POD)\nand Smooth Particle Hydrodynamics (SPH) has proven effective in various\napplications. Most MOR methods utilizing POD are implemented within a pure\nEulerian framework, while significantly less attention has been given to POD in\na Lagrangian context. In this paper, we present the POD-MOR of SPH simulations\napplied to a mass-spring-damper system with two primary objectives: 1. To\nevaluate the performance of the data-driven POD-MOR approach. 2. To investigate\npotential methods for accelerating POD-MOR computations. Although the\nmass-spring-damper system is linear, its SPH implementations are nonlinear, and\nPOD-MOR does not automatically lead to faster computations. Our findings\nindicate that (1) the POD-MOR effectively reduces the degrees of freedom in the\nSPH simulations by capturing the essential modes, and (2) in various cases, the\nacceleration of POD-MOR can be achieved without compromising accuracy. We hope\nthat our results will motivate further investigations into the design of\nPOD-MOR algorithms for nonlinear Lagrangian systems.', 'authors': ['Lidong Fang', 'Zilong Song', 'Kirk Fraser', 'Huaxiong Huang'], 'published': '2025-08-01 05:52:44+00:00', 'categories': ['physics.comp-ph'], 'url': 'http://arxiv.org/abs/2508.00335v1'}
276
2508.00475
2,508.00475
null
E2ATST: A Temporal-Spatial Optimized Energy-Efficient Architecture for Training Spiking Transformer
(1) Pengcheng Laboratory, (2) Southern University of Science and Technology, (3) Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, (4) University of Chinese Academy of Sciences
2025-08-01
{'id': '2508.00475', 'title': 'E2ATST: A Temporal-Spatial Optimized Energy-Efficient Architecture for Training Spiking Transformer', 'abstract': '(1) Pengcheng Laboratory, (2) Southern University of Science and Technology,\n(3) Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences,\n(4) University of Chinese Academy of Sciences', 'authors': ['Yunhao Ma', 'Yanyu Lin', 'Mingjing Li', 'Puli Quan', 'Chenlin Zhou', 'Wenyue Zhang', 'Zhiwei Zhong', 'Wanyi Jia', 'Xueke Zhu', 'Qingyan Meng', 'Huihui Zhou', 'Fengwei An'], 'published': '2025-08-01 09:50:40+00:00', 'categories': ['cs.AR', 'cs.NE'], 'url': 'http://arxiv.org/abs/2508.00475v1'}
277
2508.00639
2,508.00639
null
Minimum Data, Maximum Impact: 20 annotated samples for explainable lung nodule classification
Classification models that provide human-interpretable explanations enhance clinicians' trust and usability in medical image diagnosis. One research focus is the integration and prediction of pathology-related visual attributes used by radiologists alongside the diagnosis, aligning AI decision-making with clinical reasoning. Radiologists use attributes like shape and texture as established diagnostic criteria and mirroring these in AI decision-making both enhances transparency and enables explicit validation of model outputs. However, the adoption of such models is limited by the scarcity of large-scale medical image datasets annotated with these attributes. To address this challenge, we propose synthesizing attribute-annotated data using a generative model. We enhance the Diffusion Model with attribute conditioning and train it using only 20 attribute-labeled lung nodule samples from the LIDC-IDRI dataset. Incorporating its generated images into the training of an explainable model boosts performance, increasing attribute prediction accuracy by 13.4% and target prediction accuracy by 1.8% compared to training with only the small real attribute-annotated dataset. This work highlights the potential of synthetic data to overcome dataset limitations, enhancing the applicability of explainable models in medical image analysis.
2025-08-01
{'id': '2508.00639', 'title': 'Minimum Data, Maximum Impact: 20 annotated samples for explainable lung nodule classification', 'abstract': "Classification models that provide human-interpretable explanations enhance\nclinicians' trust and usability in medical image diagnosis. One research focus\nis the integration and prediction of pathology-related visual attributes used\nby radiologists alongside the diagnosis, aligning AI decision-making with\nclinical reasoning. Radiologists use attributes like shape and texture as\nestablished diagnostic criteria and mirroring these in AI decision-making both\nenhances transparency and enables explicit validation of model outputs.\nHowever, the adoption of such models is limited by the scarcity of large-scale\nmedical image datasets annotated with these attributes. To address this\nchallenge, we propose synthesizing attribute-annotated data using a generative\nmodel. We enhance the Diffusion Model with attribute conditioning and train it\nusing only 20 attribute-labeled lung nodule samples from the LIDC-IDRI dataset.\nIncorporating its generated images into the training of an explainable model\nboosts performance, increasing attribute prediction accuracy by 13.4% and\ntarget prediction accuracy by 1.8% compared to training with only the small\nreal attribute-annotated dataset. This work highlights the potential of\nsynthetic data to overcome dataset limitations, enhancing the applicability of\nexplainable models in medical image analysis.", 'authors': ['Luisa Gallée', 'Catharina Silvia Lisson', 'Christoph Gerhard Lisson', 'Daniela Drees', 'Felix Weig', 'Daniel Vogele', 'Meinrad Beer', 'Michael Götz'], 'published': '2025-08-01 13:54:34+00:00', 'categories': ['cs.CV'], 'url': 'http://arxiv.org/abs/2508.00639v1'}
278
2508.00708
2,508.00708
null
Szegő-Type Limit Theorem in the Drury-Arveson space
We state and prove a version of Szego's first limit theorem for Toeplitz-like operators on the Drury-Arveson space in the unit ball.
2025-08-01
{'id': '2508.00708', 'title': 'Szegő-Type Limit Theorem in the Drury-Arveson space', 'abstract': "We state and prove a version of Szeg\\H{o}'s first limit theorem for\nToeplitz-like operators on the Drury-Arveson space in the unit ball.", 'authors': ['Arya Gayathri Memana'], 'published': '2025-08-01 15:23:17+00:00', 'categories': ['math.FA'], 'url': 'http://arxiv.org/abs/2508.00708v1'}
279
2508.00501
2,508.00501
null
VR-PTOLEMAIC: A Virtual Environment for the Perceptual Testing of Spatial Audio Algorithms
The perceptual evaluation of spatial audio algorithms is an important step in the development of immersive audio applications, as it ensures that synthesized sound fields meet quality standards in terms of listening experience, spatial perception and auditory realism. To support these evaluations, virtual reality can offer a powerful platform by providing immersive and interactive testing environments. In this paper, we present VR-PTOLEMAIC, a virtual reality evaluation system designed for assessing spatial audio algorithms. The system implements the MUSHRA (MUlti-Stimulus test with Hidden Reference and Anchor) evaluation methodology into a virtual environment. In particular, users can position themselves in each of the 25 simulated listening positions of a virtually recreated seminar room and evaluate simulated acoustic responses with respect to the actually recorded second-order ambisonic room impulse responses, all convolved with various source signals. We evaluated the usability of the proposed framework through an extensive testing campaign in which assessors were asked to compare the reconstruction capabilities of various sound field reconstruction algorithms. Results show that the VR platform effectively supports the assessment of spatial audio algorithms, with generally positive feedback on user experience and immersivity.
2025-08-01
{'id': '2508.00501', 'title': 'VR-PTOLEMAIC: A Virtual Environment for the Perceptual Testing of Spatial Audio Algorithms', 'abstract': 'The perceptual evaluation of spatial audio algorithms is an important step in\nthe development of immersive audio applications, as it ensures that synthesized\nsound fields meet quality standards in terms of listening experience, spatial\nperception and auditory realism. To support these evaluations, virtual reality\ncan offer a powerful platform by providing immersive and interactive testing\nenvironments. In this paper, we present VR-PTOLEMAIC, a virtual reality\nevaluation system designed for assessing spatial audio algorithms. The system\nimplements the MUSHRA (MUlti-Stimulus test with Hidden Reference and Anchor)\nevaluation methodology into a virtual environment. In particular, users can\nposition themselves in each of the 25 simulated listening positions of a\nvirtually recreated seminar room and evaluate simulated acoustic responses with\nrespect to the actually recorded second-order ambisonic room impulse responses,\nall convolved with various source signals. We evaluated the usability of the\nproposed framework through an extensive testing campaign in which assessors\nwere asked to compare the reconstruction capabilities of various sound field\nreconstruction algorithms. Results show that the VR platform effectively\nsupports the assessment of spatial audio algorithms, with generally positive\nfeedback on user experience and immersivity.', 'authors': ['Paolo Ostan', 'Francesca Del Gaudio', 'Federico Miotello', 'Mirco Pezzoli', 'Fabio Antonacci'], 'published': '2025-08-01 10:26:11+00:00', 'categories': ['eess.AS', 'cs.SD'], 'url': 'http://arxiv.org/abs/2508.00501v1'}
311
2508.00260
2,508.0026
null
Instruction-Grounded Visual Projectors for Continual Learning of Generative Vision-Language Models
Continual learning enables pre-trained generative vision-language models (VLMs) to incorporate knowledge from new tasks without retraining data from previous ones. Recent methods update a visual projector to translate visual information for new tasks, connecting pre-trained vision encoders with large language models. However, such adjustments may cause the models to prioritize visual inputs over language instructions, particularly learning tasks with repetitive types of textual instructions. To address the neglect of language instructions, we propose a novel framework that grounds the translation of visual information on instructions for language models. We introduce a mixture of visual projectors, each serving as a specialized visual-to-language translation expert based on the given instruction context to adapt to new tasks. To avoid using experts for irrelevant instruction contexts, we propose an expert recommendation strategy that reuses experts for tasks similar to those previously learned. Additionally, we introduce expert pruning to alleviate interference from the use of experts that cumulatively activated in previous tasks. Extensive experiments on diverse vision-language tasks demonstrate that our method outperforms existing continual learning approaches by generating instruction-following responses.
2025-08-01
{'id': '2508.00260', 'title': 'Instruction-Grounded Visual Projectors for Continual Learning of Generative Vision-Language Models', 'abstract': 'Continual learning enables pre-trained generative vision-language models\n(VLMs) to incorporate knowledge from new tasks without retraining data from\nprevious ones. Recent methods update a visual projector to translate visual\ninformation for new tasks, connecting pre-trained vision encoders with large\nlanguage models. However, such adjustments may cause the models to prioritize\nvisual inputs over language instructions, particularly learning tasks with\nrepetitive types of textual instructions. To address the neglect of language\ninstructions, we propose a novel framework that grounds the translation of\nvisual information on instructions for language models. We introduce a mixture\nof visual projectors, each serving as a specialized visual-to-language\ntranslation expert based on the given instruction context to adapt to new\ntasks. To avoid using experts for irrelevant instruction contexts, we propose\nan expert recommendation strategy that reuses experts for tasks similar to\nthose previously learned. Additionally, we introduce expert pruning to\nalleviate interference from the use of experts that cumulatively activated in\nprevious tasks. Extensive experiments on diverse vision-language tasks\ndemonstrate that our method outperforms existing continual learning approaches\nby generating instruction-following responses.', 'authors': ['Hyundong Jin', 'Hyung Jin Chang', 'Eunwoo Kim'], 'published': '2025-08-01 02:08:09+00:00', 'categories': ['cs.CV', 'cs.MM'], 'url': 'http://arxiv.org/abs/2508.00260v1'}
280
2508.00390
2,508.0039
null
SA-GCS: Semantic-Aware Gaussian Curriculum Scheduling for UAV Vision-Language Navigation
Unmanned Aerial Vehicle (UAV) Vision-Language Navigation (VLN) aims to enable agents to accurately localize targets and plan flight paths in complex environments based on natural language instructions, with broad applications in intelligent inspection, disaster rescue, and urban monitoring. Recent progress in Vision-Language Models (VLMs) has provided strong semantic understanding for this task, while reinforcement learning (RL) has emerged as a promising post-training strategy to further improve generalization. However, existing RL methods often suffer from inefficient use of training data, slow convergence, and insufficient consideration of the difficulty variation among training samples, which limits further performance improvement. To address these challenges, we propose Semantic-Aware Gaussian Curriculum Scheduling (SA-GCS), a novel training framework that systematically integrates Curriculum Learning (CL) into RL. SA-GCS employs a Semantic-Aware Difficulty Estimator (SA-DE) to quantify the complexity of training samples and a Gaussian Curriculum Scheduler (GCS) to dynamically adjust the sampling distribution, enabling a smooth progression from easy to challenging tasks. This design significantly improves training efficiency, accelerates convergence, and enhances overall model performance. Extensive experiments on the CityNav benchmark demonstrate that SA-GCS consistently outperforms strong baselines across all metrics, achieves faster and more stable convergence, and generalizes well across models of different scales, highlighting its robustness and scalability. The implementation of our approach is publicly available.
2025-08-01
{'id': '2508.00390', 'title': 'SA-GCS: Semantic-Aware Gaussian Curriculum Scheduling for UAV Vision-Language Navigation', 'abstract': 'Unmanned Aerial Vehicle (UAV) Vision-Language Navigation (VLN) aims to enable\nagents to accurately localize targets and plan flight paths in complex\nenvironments based on natural language instructions, with broad applications in\nintelligent inspection, disaster rescue, and urban monitoring. Recent progress\nin Vision-Language Models (VLMs) has provided strong semantic understanding for\nthis task, while reinforcement learning (RL) has emerged as a promising\npost-training strategy to further improve generalization. However, existing RL\nmethods often suffer from inefficient use of training data, slow convergence,\nand insufficient consideration of the difficulty variation among training\nsamples, which limits further performance improvement. To address these\nchallenges, we propose \\textbf{Semantic-Aware Gaussian Curriculum Scheduling\n(SA-GCS)}, a novel training framework that systematically integrates Curriculum\nLearning (CL) into RL. SA-GCS employs a Semantic-Aware Difficulty Estimator\n(SA-DE) to quantify the complexity of training samples and a Gaussian\nCurriculum Scheduler (GCS) to dynamically adjust the sampling distribution,\nenabling a smooth progression from easy to challenging tasks. This design\nsignificantly improves training efficiency, accelerates convergence, and\nenhances overall model performance. Extensive experiments on the CityNav\nbenchmark demonstrate that SA-GCS consistently outperforms strong baselines\nacross all metrics, achieves faster and more stable convergence, and\ngeneralizes well across models of different scales, highlighting its robustness\nand scalability. The implementation of our approach is publicly available.', 'authors': ['Hengxing Cai', 'Jinhan Dong', 'Yijie Rao', 'Jingcheng Deng', 'Jingjun Tan', 'Qien Chen', 'Haidong Wang', 'Zhen Wang', 'Shiyu Huang', 'Agachai Sumalee', 'Renxin Zhong'], 'published': '2025-08-01 07:35:48+00:00', 'categories': ['cs.CL'], 'url': 'http://arxiv.org/abs/2508.00390v1'}
281
2508.00362
2,508.00362
null
A Whole-Body Motion Imitation Framework from Human Data for Full-Size Humanoid Robot
Motion imitation is a pivotal and effective approach for humanoid robots to achieve a more diverse range of complex and expressive movements, making their performances more human-like. However, the significant differences in kinematics and dynamics between humanoid robots and humans present a major challenge in accurately imitating motion while maintaining balance. In this paper, we propose a novel whole-body motion imitation framework for a full-size humanoid robot. The proposed method employs contact-aware whole-body motion retargeting to mimic human motion and provide initial values for reference trajectories, and the non-linear centroidal model predictive controller ensures the motion accuracy while maintaining balance and overcoming external disturbances in real time. The assistance of the whole-body controller allows for more precise torque control. Experiments have been conducted to imitate a variety of human motions both in simulation and in a real-world humanoid robot. These experiments demonstrate the capability of performing with accuracy and adaptability, which validates the effectiveness of our approach.
2025-08-01
{'id': '2508.00362', 'title': 'A Whole-Body Motion Imitation Framework from Human Data for Full-Size Humanoid Robot', 'abstract': 'Motion imitation is a pivotal and effective approach for humanoid robots to\nachieve a more diverse range of complex and expressive movements, making their\nperformances more human-like. However, the significant differences in\nkinematics and dynamics between humanoid robots and humans present a major\nchallenge in accurately imitating motion while maintaining balance. In this\npaper, we propose a novel whole-body motion imitation framework for a full-size\nhumanoid robot. The proposed method employs contact-aware whole-body motion\nretargeting to mimic human motion and provide initial values for reference\ntrajectories, and the non-linear centroidal model predictive controller ensures\nthe motion accuracy while maintaining balance and overcoming external\ndisturbances in real time. The assistance of the whole-body controller allows\nfor more precise torque control. Experiments have been conducted to imitate a\nvariety of human motions both in simulation and in a real-world humanoid robot.\nThese experiments demonstrate the capability of performing with accuracy and\nadaptability, which validates the effectiveness of our approach.', 'authors': ['Zhenghan Chen', 'Haodong Zhang', 'Dongqi Wang', 'Jiyu Yu', 'Haocheng Xu', 'Yue Wang', 'Rong Xiong'], 'published': '2025-08-01 06:45:55+00:00', 'categories': ['cs.RO'], 'url': 'http://arxiv.org/abs/2508.00362v1'}
282
2508.00224
2,508.00224
null
A Keynesian Intertemporal Synthesis (KIS) Model: Towards a unified and empirically grounded framework for fiscal policy
This paper develops a new generation of the Keynesian Intertemporal Synthesis (KIS) Model, a macroeconomic framework designed to reconcile the empirical strengths of the Post-Keynesian (PK) and New Keynesian (NK) traditions. The central innovation of this work is the abandonment of the traditional Cobb-Douglas production function in favor of a Constant Elasticity of Substitution (CES) specification. This modification is directly motivated by the compelling evidence from the meta-analysis by Gechert et al. (2021), which emphatically rejects the hypothesis of a unit elasticity of substitution between capital and labor. We integrate this finding with the conclusions from a wide range of meta-analyses on the state-dependent heterogeneity of fiscal multipliers (Gechert and Rannenberg, 2018), the productivity of public capital (Bom and Ligthart, 2014), the effectiveness hierarchy of spending instruments (Gechert, 2015), and the empirical failure of Ricardian Equivalence (Stanley, 1998). The resulting KIS-CES model, while based on intertemporal optimization, incorporates household heterogeneity, non-standard preferences that value wealth and penalize debt, and a monetary policy constrained by the zero lower bound. The mathematical derivations reveal that the elasticity of substitution, calibrated to an empirically plausible value of $< 1$, becomes a key parameter that modulates income distribution and magnifies the crowding-in effect of public investment. The model generates an endogenous MPC, a nonlinear fiscal multiplier that increases dramatically in crises, and a multiplier for public investment that is structurally higher than that for consumption, thus offering a unified, rigorous, and, above all, empirically disciplined theoretical framework.
2025-08-01
{'id': '2508.00224', 'title': 'A Keynesian Intertemporal Synthesis (KIS) Model: Towards a unified and empirically grounded framework for fiscal policy', 'abstract': 'This paper develops a new generation of the Keynesian Intertemporal Synthesis\n(KIS) Model, a macroeconomic framework designed to reconcile the empirical\nstrengths of the Post-Keynesian (PK) and New Keynesian (NK) traditions. The\ncentral innovation of this work is the abandonment of the traditional\nCobb-Douglas production function in favor of a Constant Elasticity of\nSubstitution (CES) specification. This modification is directly motivated by\nthe compelling evidence from the meta-analysis by Gechert et al. (2021), which\nemphatically rejects the hypothesis of a unit elasticity of substitution\nbetween capital and labor. We integrate this finding with the conclusions from\na wide range of meta-analyses on the state-dependent heterogeneity of fiscal\nmultipliers (Gechert and Rannenberg, 2018), the productivity of public capital\n(Bom and Ligthart, 2014), the effectiveness hierarchy of spending instruments\n(Gechert, 2015), and the empirical failure of Ricardian Equivalence (Stanley,\n1998). The resulting KIS-CES model, while based on intertemporal optimization,\nincorporates household heterogeneity, non-standard preferences that value\nwealth and penalize debt, and a monetary policy constrained by the zero lower\nbound. The mathematical derivations reveal that the elasticity of substitution,\ncalibrated to an empirically plausible value of $\\sigma < 1$, becomes a key\nparameter that modulates income distribution and magnifies the crowding-in\neffect of public investment. The model generates an endogenous MPC, a nonlinear\nfiscal multiplier that increases dramatically in crises, and a multiplier for\npublic investment that is structurally higher than that for consumption, thus\noffering a unified, rigorous, and, above all, empirically disciplined\ntheoretical framework.', 'authors': ['Ricardo Alonzo Fernández Salguero'], 'published': '2025-08-01 00:03:20+00:00', 'categories': ['econ.TH'], 'url': 'http://arxiv.org/abs/2508.00224v1'}
283
2508.00671
2,508.00671
null
Chemical abundances of seven stars in the GD-1 stream
We present the first detailed chemical abundances for seven GD-1 stream stars from Subaru/HDS spectroscopy. Atmospheric parameters were derived via color calibrations ($T_eff$) and iterative spectroscopic analysis. LTE abundances for 14 elements ($$, odd-Z, iron-peak, n-capture) were measured. Six stars trace the main orbit, one resides in a `blob'. All exhibit tightly clustered metallicities ([Fe/H] = -2.38, intrinsic dispersion smaller than 0.05 dex, average uncertainty is about 0.13 dex). While one star shows binary mass transfer signatures, the other six display consistent abundance patterns (dispersions $<$ uncertainties). Their iron-peak elements (Sc, Cr, Mn, Ni) match Milky Way halo stars. In contrast, Y and Sr are systematically lower than halo stars of similar [Fe/H]. Significantly, six stars show consistently enhanced [Eu/Fe] $$ 0.60 ($$ = 0.08). A tight Ba-Eu correlation (r = 0.83, p=0.04) exists, with [Ba/Fe] = -0.03 $$ 0.05, indicating a common r-process origin. This extreme chemical homogeneity strongly supports an origin from a single disrupted globular cluster. The lack of light-element anti-correlations may stem from our sample size or the progenitor's low mass.
2025-08-01
{'id': '2508.00671', 'title': 'Chemical abundances of seven stars in the GD-1 stream', 'abstract': "We present the first detailed chemical abundances for seven GD-1 stream stars\nfrom Subaru/HDS spectroscopy. Atmospheric parameters were derived via color\ncalibrations ($T\\rm_{eff}$) and iterative spectroscopic analysis. LTE\nabundances for 14 elements ($\\alpha$, odd-Z, iron-peak, n-capture) were\nmeasured. Six stars trace the main orbit, one resides in a `blob'. All exhibit\ntightly clustered metallicities ([Fe/H] = -2.38, {\\bf intrinsic dispersion\nsmaller than 0.05 dex, average uncertainty is about 0.13 dex}). While one star\nshows binary mass transfer signatures, the other six display consistent\nabundance patterns (dispersions $<$ uncertainties). Their iron-peak elements\n(Sc, Cr, Mn, Ni) match Milky Way halo stars. In contrast, Y and Sr are\nsystematically lower than halo stars of similar [Fe/H]. Significantly, six\nstars show consistently enhanced [Eu/Fe] $\\sim$ 0.60 ($\\sigma$ = 0.08). A tight\nBa-Eu correlation (r = 0.83, p=0.04) exists, with [Ba/Fe] = -0.03 $\\pm$ 0.05,\nindicating a common r-process origin. This extreme chemical homogeneity\nstrongly supports an origin from a single disrupted globular cluster. The lack\nof light-element anti-correlations may stem from our sample size or the\nprogenitor's low mass.", 'authors': ['Jing-Kun Zhao', 'Guang-Wei Li', 'Wako Aoki', 'Gang Zhao', 'Guo-Chao Yang', 'Jian-Rong Shi', 'Hai-Ning Li', 'Tadafumi Matsuno', 'Miho Ishigaki', 'Takuma Suda', 'Satoshi Honda', 'Yu-Qin Chen', 'Qian-Fan Xing', 'Hong-Liang Yan', 'Yong Yang', 'Xian-Hao Ye'], 'published': '2025-08-01 14:46:09+00:00', 'categories': ['astro-ph.GA'], 'url': 'http://arxiv.org/abs/2508.00671v1'}
289
2508.00593
2,508.00593
null
Can User Feedback Help Issue Detection? An Empirical Study on a One-billion-user Online Service System
Background: It has long been suggested that user feedback, typically written in natural language by end-users, can help issue detection. However, for large-scale online service systems that receive a tremendous amount of feedback, it remains a challenging task to identify severe issues from user feedback. Aims: To develop a better feedback-based issue detection approach, it is crucial first to gain a comprehensive understanding of the characteristics of user feedback in real production systems. Method: In this paper, we conduct an empirical study on 50,378,766 user feedback items from six real-world services in a one-billion-user online service system. We first study what users provide in their feedback. We then examine whether certain features of feedback items can be good indicators of severe issues. Finally, we investigate whether adopting machine learning techniques to analyze user feedback is reasonable. Results: Our results show that a large proportion of user feedback provides irrelevant information about system issues. As a result, it is crucial to filter out issue-irrelevant information when processing user feedback. Moreover, we find severe issues that cannot be easily detected based solely on user feedback characteristics. Finally, we find that the distributions of the feedback topics in different time intervals are similar. This confirms that designing machine learning-based approaches is a viable direction for better analyzing user feedback. Conclusions: We consider that our findings can serve as an empirical foundation for feedback-based issue detection in large-scale service systems, which sheds light on the design and implementation of practical issue detection approaches.
2025-08-01
{'id': '2508.00593', 'title': 'Can User Feedback Help Issue Detection? An Empirical Study on a One-billion-user Online Service System', 'abstract': 'Background: It has long been suggested that user feedback, typically written\nin natural language by end-users, can help issue detection. However, for\nlarge-scale online service systems that receive a tremendous amount of\nfeedback, it remains a challenging task to identify severe issues from user\nfeedback. Aims: To develop a better feedback-based issue detection approach, it\nis crucial first to gain a comprehensive understanding of the characteristics\nof user feedback in real production systems. Method: In this paper, we conduct\nan empirical study on 50,378,766 user feedback items from six real-world\nservices in a one-billion-user online service system. We first study what users\nprovide in their feedback. We then examine whether certain features of feedback\nitems can be good indicators of severe issues. Finally, we investigate whether\nadopting machine learning techniques to analyze user feedback is reasonable.\nResults: Our results show that a large proportion of user feedback provides\nirrelevant information about system issues. As a result, it is crucial to\nfilter out issue-irrelevant information when processing user feedback.\nMoreover, we find severe issues that cannot be easily detected based solely on\nuser feedback characteristics. Finally, we find that the distributions of the\nfeedback topics in different time intervals are similar. This confirms that\ndesigning machine learning-based approaches is a viable direction for better\nanalyzing user feedback. Conclusions: We consider that our findings can serve\nas an empirical foundation for feedback-based issue detection in large-scale\nservice systems, which sheds light on the design and implementation of\npractical issue detection approaches.', 'authors': ['Shuyao Jiang', 'Jiazhen Gu', 'Wujie Zheng', 'Yangfan Zhou', 'Michael R. Lyu'], 'published': '2025-08-01 12:49:07+00:00', 'categories': ['cs.SE'], 'url': 'http://arxiv.org/abs/2508.00593v1'}
284
2508.00618
2,508.00618
null
Locally non-trivial fibred surfaces with maximal unitary rank
Let $fSB$ a locally non-trivial fibred surface with fibres of genus $g$. Let $u_f$ be its unitary rank, i.e. the rank of the flat unitary part in the second Fujita decomposition. We study in detail the case when $u_f$ is maximal, i.e. $u_f=g-1$. In this case necessarily $g6$, but examples in genus $5$ and $6$ are not known, and conjecturally do not exist. We prove a strong slope inequality for these extremal cases. We then use this inequality, together with results on trigonal curves, to give new constraints on the case $g=6$, $u_f=5$. In particular, we prove that the index of the surface is always strictly positive and give strong limitations on the possible classes of the relative canonical divisor
2025-08-01
{'id': '2508.00618', 'title': 'Locally non-trivial fibred surfaces with maximal unitary rank', 'abstract': 'Let $f\\colon S\\to B$ a locally non-trivial fibred surface with fibres of\ngenus $g$. Let $u_f$ be its unitary rank, i.e. the rank of the flat unitary\npart in the second Fujita decomposition. We study in detail the case when $u_f$\nis maximal, i.e. $u_f=g-1$. In this case necessarily $g\\leq 6$, but examples in\ngenus $5$ and $6$ are not known, and conjecturally do not exist. We prove a\nstrong slope inequality for these extremal cases. We then use this inequality,\ntogether with results on trigonal curves, to give new constraints on the case\n$g=6$, $u_f=5$. In particular, we prove that the index of the surface is always\nstrictly positive and give strong limitations on the possible classes of the\nrelative canonical divisor', 'authors': ['Lidia Stoppino'], 'published': '2025-08-01 13:27:26+00:00', 'categories': ['math.AG', '14J10, 14D06, 14D07, 14G35, 14H40'], 'url': 'http://arxiv.org/abs/2508.00618v1'}
285
2508.00560
2,508.0056
null
Wave-mixing cathodoluminescence microscopy of low-frequency excitations
Nonlinear optical phenomena such as parametric amplification and frequency conversion are typically driven by external optical fields. Free electrons can also act as electromagnetic sources, offering unmatched spatial precision. Combining optical and electron-induced fields via the nonlinear response of material structures therefore holds potential for revealing new physical phenomena and enabling disruptive applications. Here, we theoretically investigate wave mixing between external light and the evanescent fields of free electrons, giving rise to inelastic photon scattering mediated by the second-order nonlinear response of a specimen. Specifically, an incident photon may be blue- or red-shifted, while the passing electron correspondingly loses or gains energy. These processes are strongly enhanced when the frequency shift matches an optical resonance of the specimen. We present a general theoretical framework to quantify the photon conversion probability and demonstrate its application by revealing far-infrared vibrational fingerprints of retinal using only visible light. Beyond its fundamental interest, this phenomenon offers a practical approach for spatially mapping low-frequency excitations with nanometer resolution using visible photon energies and existing electron microscopes.
2025-08-01
{'id': '2508.00560', 'title': 'Wave-mixing cathodoluminescence microscopy of low-frequency excitations', 'abstract': 'Nonlinear optical phenomena such as parametric amplification and frequency\nconversion are typically driven by external optical fields. Free electrons can\nalso act as electromagnetic sources, offering unmatched spatial precision.\nCombining optical and electron-induced fields via the nonlinear response of\nmaterial structures therefore holds potential for revealing new physical\nphenomena and enabling disruptive applications. Here, we theoretically\ninvestigate wave mixing between external light and the evanescent fields of\nfree electrons, giving rise to inelastic photon scattering mediated by the\nsecond-order nonlinear response of a specimen. Specifically, an incident photon\nmay be blue- or red-shifted, while the passing electron correspondingly loses\nor gains energy. These processes are strongly enhanced when the frequency shift\nmatches an optical resonance of the specimen. We present a general theoretical\nframework to quantify the photon conversion probability and demonstrate its\napplication by revealing far-infrared vibrational fingerprints of retinal using\nonly visible light. Beyond its fundamental interest, this phenomenon offers a\npractical approach for spatially mapping low-frequency excitations with\nnanometer resolution using visible photon energies and existing electron\nmicroscopes.', 'authors': ['Leila Prelat', 'Eduardo J. C. Dias', 'F. Javier García de Abajo'], 'published': '2025-08-01 12:01:09+00:00', 'categories': ['cond-mat.mes-hall'], 'url': 'http://arxiv.org/abs/2508.00560v1'}
286
2508.00486
2,508.00486
null
The Bose-Hubbard polaron from weak to strong coupling
We investigate the zero-temperature properties of a mobile impurity immersed in a bath of bosonic particles confined to a square lattice. We analyze the regimes of attractive and repulsive coupling between the impurity and the bath particles for different strengths of boson-boson interactions in the bath, using exact large-scale quantum Monte-Carlo simulations in the grand canonical ensemble. For weak coupling, the polaron mass ratio is found to decrease around the Mott insulator (MI) to superfluid (SF) transition of the bath, as predicted by recent theory, confirming the possible use of the impurity as a probe for the transition. For strong coupling in the MI regime, instead, the impurity is found to modify the bath density by binding to an extra bath particle or a hole, depending on the sign of the polaron-bath interactions. While the binding prevent the aforementioned use of the polaron mass ratio as an MI-SF transition probe, we show that it can be used instead as a probe of the binding itself. Our exact numerical results provide a benchmark for comparing lattice Bose polaron theories and are relevant for experiments with cold atoms trapped in optical lattices, where the presence of a confining harmonic potential can be modeled by a slowly varying local chemical potential.
2025-08-01
{'id': '2508.00486', 'title': 'The Bose-Hubbard polaron from weak to strong coupling', 'abstract': 'We investigate the zero-temperature properties of a mobile impurity immersed\nin a bath of bosonic particles confined to a square lattice. We analyze the\nregimes of attractive and repulsive coupling between the impurity and the bath\nparticles for different strengths of boson-boson interactions in the bath,\nusing exact large-scale quantum Monte-Carlo simulations in the grand canonical\nensemble. For weak coupling, the polaron mass ratio is found to decrease around\nthe Mott insulator (MI) to superfluid (SF) transition of the bath, as predicted\nby recent theory, confirming the possible use of the impurity as a probe for\nthe transition. For strong coupling in the MI regime, instead, the impurity is\nfound to modify the bath density by binding to an extra bath particle or a\nhole, depending on the sign of the polaron-bath interactions. While the binding\nprevent the aforementioned use of the polaron mass ratio as an MI-SF transition\nprobe, we show that it can be used instead as a probe of the binding itself.\nOur exact numerical results provide a benchmark for comparing lattice Bose\npolaron theories and are relevant for experiments with cold atoms trapped in\noptical lattices, where the presence of a confining harmonic potential can be\nmodeled by a slowly varying local chemical potential.', 'authors': ['Tom Hartweg', 'Tanul Gupta', 'Guido Pupillo'], 'published': '2025-08-01 10:05:05+00:00', 'categories': ['cond-mat.quant-gas', 'quant-ph'], 'url': 'http://arxiv.org/abs/2508.00486v1'}
287
2508.00437
2,508.00437
null
Replacement-Type Quantum Gates
We introduce the paradigm of replacement-type quantum gates. This type of gate introduces input qubits, candidate qubits, and output qubits. The candidate qubits are prepared such, that a displacement conditional on the input qubit results in the targeted output state. Finally, the circuit continues with the output qubits constructed from the candidate qubits instead of the input qubits, thus the name "replacement-type gate". We present examples of replacement-type $X$ and $CNOT$ gates realized with spin qubits and with neutral atom qubits. By making use of the extended Hilbert space, including the position of the particles, these gates approximately preserve the innate noise bias of the qubits. The gate preserves the noise-bias which motivates advanced quantum computer architectures with error correction.
2025-08-01
{'id': '2508.00437', 'title': 'Replacement-Type Quantum Gates', 'abstract': 'We introduce the paradigm of replacement-type quantum gates. This type of\ngate introduces input qubits, candidate qubits, and output qubits. The\ncandidate qubits are prepared such, that a displacement conditional on the\ninput qubit results in the targeted output state. Finally, the circuit\ncontinues with the output qubits constructed from the candidate qubits instead\nof the input qubits, thus the name "replacement-type gate". We present examples\nof replacement-type $X$ and $\\mathrm{CNOT}$ gates realized with spin qubits and\nwith neutral atom qubits. By making use of the extended Hilbert space,\nincluding the position of the particles, these gates approximately preserve the\ninnate noise bias of the qubits. The gate preserves the noise-bias which\nmotivates advanced quantum computer architectures with error correction.', 'authors': ['Florian Ginzel', 'Javad Kazemi', 'Valentin Torggler', 'Wolfgang Lechner'], 'published': '2025-08-01 08:51:45+00:00', 'categories': ['quant-ph', 'cond-mat.mes-hall'], 'url': 'http://arxiv.org/abs/2508.00437v1'}
288
2508.00269
2,508.00269
null
chipfiring: A Python Package for Efficient Mathematical Analysis of Chip-Firing Games on Multigraphs
This paper presents `chipfiring`, a comprehensive Python package for the mathematical analysis of chip-firing games on finite graphs. The package provides a robust toolkit for defining graphs and chip configurations (divisors), performing chip-firing operations, and analyzing fundamental properties such as winnability, linear equivalence, and divisor rank. We detail the core components of the library, including its object-oriented graph and divisor implementations, integrated Laplacian matrix computations, and an efficient implementation of Dhar's algorithm for determining the solvability of the dollar game. The `chipfiring` package is designed for researchers and students in graph theory, combinatorics, and algebraic geometry, providing essential algorithms and data structures for exploring these rich mathematical models. We describe the library's architecture, illustrate its usage with comprehensive examples, and highlight its specialized contributions compared to general-purpose graph libraries.
2025-08-01
{'id': '2508.00269', 'title': 'chipfiring: A Python Package for Efficient Mathematical Analysis of Chip-Firing Games on Multigraphs', 'abstract': "This paper presents `chipfiring`, a comprehensive Python package for the\nmathematical analysis of chip-firing games on finite graphs. The package\nprovides a robust toolkit for defining graphs and chip configurations\n(divisors), performing chip-firing operations, and analyzing fundamental\nproperties such as winnability, linear equivalence, and divisor rank. We detail\nthe core components of the library, including its object-oriented graph and\ndivisor implementations, integrated Laplacian matrix computations, and an\nefficient implementation of Dhar's algorithm for determining the solvability of\nthe dollar game. The `chipfiring` package is designed for researchers and\nstudents in graph theory, combinatorics, and algebraic geometry, providing\nessential algorithms and data structures for exploring these rich mathematical\nmodels. We describe the library's architecture, illustrate its usage with\ncomprehensive examples, and highlight its specialized contributions compared to\ngeneral-purpose graph libraries.", 'authors': ['Dhyey Dharmendrakumar Mavani', 'Tairan Ji', 'Nathan Pflueger'], 'published': '2025-08-01 02:29:20+00:00', 'categories': ['math.CO', 'cs.CG', 'cs.MS', 'math.AG'], 'url': 'http://arxiv.org/abs/2508.00269v2'}
290
2508.00642
2,508.00642
null
Prototype Development and Calibration of the CUbesat Solar Polarimeter (CUSP)
The space-based CUbesat Solar Polarimeter (CUSP) mission aims to measure the linear polarization of solar flares in the hard X-ray band by means of a Compton scattering polarimeter. CUSP will allow to study the magnetic reconnection and particle acceleration in the flaring magnetic structures of our star with its unprecedented sensitivity to solar flare polarization. CUSP is a project in the framework of the Alcor Program of the Italian Space Agency aimed to develop new CubeSat missions. It has been proposed as a constellation of a two Cubesat mission to monitor the Sun for Space Weather, and will proceed with a single-satellite asset in its baseline implementation. In the frame of CUSP's Phase B study, that started in December 2024 for a 1-year period, we present the development status of this dual-phase polarimeter. Preliminary laboratory results using two chains of acquisition will be discussed. The first chain of acquisition, based on the Hamamatsu R7600 multi-anode photomultiplier tubes coupled to plastic scintillator bars and read out by the MAROC-3A ASIC, is used to detect the Compton scattering of incoming photons. On the other hand, GAGG crystals coupled to avalanche photo-diodes with a readout based on the SKIROC-2A ASIC are used to absorb the scattered photons. By reconstructing the azimuthal scattering direction for many incoming photons, one can infer the linear polarization degree and angle of the source. We will discuss the calibration results obtained with our prototype detector by using well-known radioactive isotopes, allowing us to assess the performances of our detector over the full 25-100 keV energy range.
2025-08-01
{'id': '2508.00642', 'title': 'Prototype Development and Calibration of the CUbesat Solar Polarimeter (CUSP)', 'abstract': "The space-based CUbesat Solar Polarimeter (CUSP) mission aims to measure the\nlinear polarization of solar flares in the hard X-ray band by means of a\nCompton scattering polarimeter. CUSP will allow to study the magnetic\nreconnection and particle acceleration in the flaring magnetic structures of\nour star with its unprecedented sensitivity to solar flare polarization. CUSP\nis a project in the framework of the Alcor Program of the Italian Space Agency\naimed to develop new CubeSat missions. It has been proposed as a constellation\nof a two Cubesat mission to monitor the Sun for Space Weather, and will proceed\nwith a single-satellite asset in its baseline implementation.\n In the frame of CUSP's Phase B study, that started in December 2024 for a\n1-year period, we present the development status of this dual-phase\npolarimeter. Preliminary laboratory results using two chains of acquisition\nwill be discussed. The first chain of acquisition, based on the Hamamatsu R7600\nmulti-anode photomultiplier tubes coupled to plastic scintillator bars and read\nout by the MAROC-3A ASIC, is used to detect the Compton scattering of incoming\nphotons. On the other hand, GAGG crystals coupled to avalanche photo-diodes\nwith a readout based on the SKIROC-2A ASIC are used to absorb the scattered\nphotons. By reconstructing the azimuthal scattering direction for many incoming\nphotons, one can infer the linear polarization degree and angle of the source.\nWe will discuss the calibration results obtained with our prototype detector by\nusing well-known radioactive isotopes, allowing us to assess the performances\nof our detector over the full 25-100 keV energy range.", 'authors': ['Nicolas De Angelis', 'Abhay Kumar', 'Sergio Fabiani', 'Ettore Del Monte', 'Enrico Costa', 'Giovanni Lombardi', 'Paolo Soffitta', 'Andrea Alimenti', 'Riccardo Campana', 'Mauro Centrone', 'Giovanni De Cesare', 'Sergio Di Cosimo', 'Giuseppe Di Persio', 'Alessandro Lacerenza', 'Pasqualino Loffredo', 'Gabriele Minervini', 'Fabio Muleri', 'Paolo Romano', 'Alda Rubini', 'Emanuele Scalise', 'Enrico Silva', 'Davide Albanesi', 'Ilaria Baffo', 'Daniele Brienza', 'Valerio Campamaggiore', 'Giovanni Cucinella', 'Andrea Curatolo', 'Giulia de Iulis', 'Andrea Del Re', 'Vito Di Bari', 'Simone Di Filippo', 'Immacolata Donnarumma', 'Pierluigi Fanelli', 'Nicolas Gagliardi', 'Paolo Leonetti', 'Matteo Merge', 'Dario Modenini', 'Andrea Negri', 'Daniele Pecorella', 'Massimo Perelli', 'Alice Ponti', 'Francesca Sbop', 'Paolo Tortora', 'Alessandro Turchi', 'Valerio Vagelli', 'Emanuele Zaccagnino', 'Alessandro Zambardi', 'Costantino Zazza'], 'published': '2025-08-01 13:56:39+00:00', 'categories': ['astro-ph.IM', 'astro-ph.SR', 'physics.space-ph'], 'url': 'http://arxiv.org/abs/2508.00642v1'}
291
2508.00690
2,508.0069
null
Analogues of the Milky Way-Sagittarius interaction in the TNG50: effect on the Milky Way
The Sagittarius Dwarf Galaxy is undoubtedly being disrupted in the tidal field of the Milky Way. The Milky Way disc is also found to be in a state of disequilibrium. The role of the Sagittarius in driving or contributing to this disequilibrium has been extensively investigated. Most of these studies, however, assume an initially near-equilibrium disc. It was also hypothesized that the passage of Sagittarius could have increased the star-forming activity in the Solar Neighbourhood. We check whether galaxies that have undergone cosmological evolution are affected by interactions analogous to those between the Sagittarius and the Milky Way. We use the high-resolution simulation TNG50 to look for pairs similar to the Milky Way and Sagittarius. We search within redshift z=1-0 for discs from the MW/M31 sample that interacted with a satellite more massive than 10 billion Msun, had a pericenter smaller than 50 kpc, and was on an approximately polar orbit. We exclude cases where, within 1 Gyr of the pericenter, a similar interaction occurred. In 90 percent of cases, a passage of the Sagittarius analogue had no significant effect on either the vertical velocity field of the disc or the star formation history. A response in vertical stellar kinematics can be found mostly in cold discs and mildly correlates with the strength of interactions. For star formation, the studied interactions had an effect only when little to no star formation was ongoing prior to the interaction, often due to previously disturbed star-forming discs, e.g., from AGN activity. Our results indicate that stellar discs in TNG50 are frequently vertically perturbed preceding pericenter passages of Sagittarius analogues. Future studies using other simulations and extragalactic surveys will help establish whether vertical disequilibrium is a common feature of disc galaxies or an artifact of the specific setup studied.
2025-08-01
{'id': '2508.00690', 'title': 'Analogues of the Milky Way-Sagittarius interaction in the TNG50: effect on the Milky Way', 'abstract': 'The Sagittarius Dwarf Galaxy is undoubtedly being disrupted in the tidal\nfield of the Milky Way. The Milky Way disc is also found to be in a state of\ndisequilibrium. The role of the Sagittarius in driving or contributing to this\ndisequilibrium has been extensively investigated. Most of these studies,\nhowever, assume an initially near-equilibrium disc. It was also hypothesized\nthat the passage of Sagittarius could have increased the star-forming activity\nin the Solar Neighbourhood. We check whether galaxies that have undergone\ncosmological evolution are affected by interactions analogous to those between\nthe Sagittarius and the Milky Way. We use the high-resolution simulation TNG50\nto look for pairs similar to the Milky Way and Sagittarius. We search within\nredshift z=1-0 for discs from the MW/M31 sample that interacted with a\nsatellite more massive than 10 billion Msun, had a pericenter smaller than 50\nkpc, and was on an approximately polar orbit. We exclude cases where, within 1\nGyr of the pericenter, a similar interaction occurred. In 90 percent of cases,\na passage of the Sagittarius analogue had no significant effect on either the\nvertical velocity field of the disc or the star formation history. A response\nin vertical stellar kinematics can be found mostly in cold discs and mildly\ncorrelates with the strength of interactions. For star formation, the studied\ninteractions had an effect only when little to no star formation was ongoing\nprior to the interaction, often due to previously disturbed star-forming discs,\ne.g., from AGN activity. Our results indicate that stellar discs in TNG50 are\nfrequently vertically perturbed preceding pericenter passages of Sagittarius\nanalogues. Future studies using other simulations and extragalactic surveys\nwill help establish whether vertical disequilibrium is a common feature of disc\ngalaxies or an artifact of the specific setup studied.', 'authors': ['Marcin Semczuk', 'Teresa Antoja', 'Alexandra Girón-Soto', 'Chervin F. P. Laporte'], 'published': '2025-08-01 15:06:39+00:00', 'categories': ['astro-ph.GA'], 'url': 'http://arxiv.org/abs/2508.00690v1'}
292
2508.00532
2,508.00532
null
Utilizing Deep Learning for Enhanced Tritium Detection in CCDs
This study explores the use of charge-coupled devices (CCDs) for detecting low-energy beta particles from tritium decay - a critical signal for nuclear safety, nuclear nonproliferation, and environmental monitoring. We employ a dual approach utilizing both measured CCD data and detailed Geant4 simulations. Our analysis compares classical techniques with advanced deep learning methods, including convolutional neural networks (CNNs), autoencoders trained exclusively on tritium data, and preliminary studies on boosted decision trees (BDTs). The CNN, trained on mixed signal/background datasets, demonstrates superior classification performance, while the autoencoder shows the potential of unsupervised, background-agnostic strategies. These results highlight the excellent sensitivity achievable thanks to the background rejection made possible by information-rich CCD data, paving the way for improved portable tritium monitoring.
2025-08-01
{'id': '2508.00532', 'title': 'Utilizing Deep Learning for Enhanced Tritium Detection in CCDs', 'abstract': 'This study explores the use of charge-coupled devices (CCDs) for detecting\nlow-energy beta particles from tritium decay - a critical signal for nuclear\nsafety, nuclear nonproliferation, and environmental monitoring. We employ a\ndual approach utilizing both measured CCD data and detailed Geant4 simulations.\nOur analysis compares classical techniques with advanced deep learning methods,\nincluding convolutional neural networks (CNNs), autoencoders trained\nexclusively on tritium data, and preliminary studies on boosted decision trees\n(BDTs). The CNN, trained on mixed signal/background datasets, demonstrates\nsuperior classification performance, while the autoencoder shows the potential\nof unsupervised, background-agnostic strategies. These results highlight the\nexcellent sensitivity achievable thanks to the background rejection made\npossible by information-rich CCD data, paving the way for improved portable\ntritium monitoring.', 'authors': ['E. Rofors', 'R. Heller', 'R. J. Cooper', 'J. Estrada', 'G. Moroni', 'B. Nachman', 'K. Spears'], 'published': '2025-08-01 11:13:30+00:00', 'categories': ['physics.ins-det'], 'url': 'http://arxiv.org/abs/2508.00532v1'}
293
2508.00615
2,508.00615
null
Similarity-Based Self-Construct Graph Model for Predicting Patient Criticalness Using Graph Neural Networks and EHR Data
Accurately predicting the criticalness of ICU patients (such as in-ICU mortality risk) is vital for early intervention in critical care. However, conventional models often treat each patient in isolation and struggle to exploit the relational structure in Electronic Health Records (EHR). We propose a Similarity-Based Self-Construct Graph Model (SBSCGM) that dynamically builds a patient similarity graph from multi-modal EHR data, and a HybridGraphMedGNN architecture that operates on this graph to predict patient mortality and a continuous criticalness score. SBSCGM uses a hybrid similarity measure (combining feature-based and structural similarities) to connect patients with analogous clinical profiles in real-time. The HybridGraphMedGNN integrates Graph Convolutional Network (GCN), GraphSAGE, and Graph Attention Network (GAT) layers to learn robust patient representations, leveraging both local and global graph patterns. In experiments on 6,000 ICU stays from the MIMIC-III dataset, our model achieves state-of-the-art performance (AUC-ROC $0.94$) outperforming baseline classifiers and single-type GNN models. We also demonstrate improved precision/recall and show that the attention mechanism provides interpretable insights into model predictions. Our framework offers a scalable and interpretable solution for critical care risk prediction, with potential to support clinicians in real-world ICU deployment.
2025-08-01
{'id': '2508.00615', 'title': 'Similarity-Based Self-Construct Graph Model for Predicting Patient Criticalness Using Graph Neural Networks and EHR Data', 'abstract': 'Accurately predicting the criticalness of ICU patients (such as in-ICU\nmortality risk) is vital for early intervention in critical care. However,\nconventional models often treat each patient in isolation and struggle to\nexploit the relational structure in Electronic Health Records (EHR). We propose\na Similarity-Based Self-Construct Graph Model (SBSCGM) that dynamically builds\na patient similarity graph from multi-modal EHR data, and a HybridGraphMedGNN\narchitecture that operates on this graph to predict patient mortality and a\ncontinuous criticalness score. SBSCGM uses a hybrid similarity measure\n(combining feature-based and structural similarities) to connect patients with\nanalogous clinical profiles in real-time. The HybridGraphMedGNN integrates\nGraph Convolutional Network (GCN), GraphSAGE, and Graph Attention Network (GAT)\nlayers to learn robust patient representations, leveraging both local and\nglobal graph patterns. In experiments on 6,000 ICU stays from the MIMIC-III\ndataset, our model achieves state-of-the-art performance (AUC-ROC $0.94$)\noutperforming baseline classifiers and single-type GNN models. We also\ndemonstrate improved precision/recall and show that the attention mechanism\nprovides interpretable insights into model predictions. Our framework offers a\nscalable and interpretable solution for critical care risk prediction, with\npotential to support clinicians in real-world ICU deployment.', 'authors': ['Mukesh Kumar Sahu', 'Pinki Roy'], 'published': '2025-08-01 13:25:04+00:00', 'categories': ['cs.LG', 'cs.AI'], 'url': 'http://arxiv.org/abs/2508.00615v1'}
294
2508.00589
2,508.00589
null
Context-based Motion Retrieval using Open Vocabulary Methods for Autonomous Driving
Autonomous driving systems must operate reliably in safety-critical scenarios, particularly those involving unusual or complex behavior by Vulnerable Road Users (VRUs). Identifying these edge cases in driving datasets is essential for robust evaluation and generalization, but retrieving such rare human behavior scenarios within the long tail of large-scale datasets is challenging. To support targeted evaluation of autonomous driving systems in diverse, human-centered scenarios, we propose a novel context-aware motion retrieval framework. Our method combines Skinned Multi-Person Linear (SMPL)-based motion sequences and corresponding video frames before encoding them into a shared multimodal embedding space aligned with natural language. Our approach enables the scalable retrieval of human behavior and their context through text queries. This work also introduces our dataset WayMoCo, an extension of the Waymo Open Dataset. It contains automatically labeled motion and scene context descriptions derived from generated pseudo-ground-truth SMPL sequences and corresponding image data. Our approach outperforms state-of-the-art models by up to 27.5% accuracy in motion-context retrieval, when evaluated on the WayMoCo dataset.
2025-08-01
{'id': '2508.00589', 'title': 'Context-based Motion Retrieval using Open Vocabulary Methods for Autonomous Driving', 'abstract': 'Autonomous driving systems must operate reliably in safety-critical\nscenarios, particularly those involving unusual or complex behavior by\nVulnerable Road Users (VRUs). Identifying these edge cases in driving datasets\nis essential for robust evaluation and generalization, but retrieving such rare\nhuman behavior scenarios within the long tail of large-scale datasets is\nchallenging. To support targeted evaluation of autonomous driving systems in\ndiverse, human-centered scenarios, we propose a novel context-aware motion\nretrieval framework. Our method combines Skinned Multi-Person Linear\n(SMPL)-based motion sequences and corresponding video frames before encoding\nthem into a shared multimodal embedding space aligned with natural language.\nOur approach enables the scalable retrieval of human behavior and their context\nthrough text queries. This work also introduces our dataset WayMoCo, an\nextension of the Waymo Open Dataset. It contains automatically labeled motion\nand scene context descriptions derived from generated pseudo-ground-truth SMPL\nsequences and corresponding image data. Our approach outperforms\nstate-of-the-art models by up to 27.5% accuracy in motion-context retrieval,\nwhen evaluated on the WayMoCo dataset.', 'authors': ['Stefan Englmeier', 'Max A. Büttner', 'Katharina Winter', 'Fabian B. Flohr'], 'published': '2025-08-01 12:41:52+00:00', 'categories': ['cs.CV', 'cs.CL', 'cs.IR', 'cs.RO', '68T45, 68P20, 68T10, 68T50, 68T07, 68T40', 'I.2.10; I.4.8; I.2.9; H.3.3'], 'url': 'http://arxiv.org/abs/2508.00589v1'}
295
2508.00386
2,508.00386
null
Noise Reduction Method for Radio Astronomy Single Station Observation Based on Wavelet Transform and Mathematical Morphology
The 21 cm radiation of neutral hydrogen provides crucial information for studying the early universe and its evolution. To advance this research, countries have made significant investments in constructing large low-frequency radio telescope arrays, such as the Low Frequency Array (LOFAR) and the Square Kilometre Array Phase 1 Low Frequency (SKA1-low). These instruments are pivotal for radio astronomy research. However, challenges such as ionospheric plasma interference, ambient radio noise, and instrument-related effects have become increasingly prominent, posing major obstacles in cosmology research. To address these issues, this paper proposes an efficient signal processing method that combines wavelet transform and mathematical morphology. The method involves the following steps: Background Subtraction: Background interference in radio observation signals is eliminated. Wavelet Transform: The signal, after removing background noise, undergoes a two-dimensional discrete wavelet transform. Threshold processing is then applied to the wavelet coefficients to effectively remove interference components. Wavelet Inversion: The processed signal is reconstructed using wavelet inversion. Mathematical Morphology: The reconstructed signal is further optimized using mathematical morphology to refine the results. Experimental verification was conducted using solar observation data from the Xinjiang Observatory and the Yunnan Observatory. The results demonstrate that this method successfully removes interference signals while preserving useful signals, thus improving the accuracy of radio astronomy observations and reducing the impact of radio frequency interference (RFI).
2025-08-01
{'id': '2508.00386', 'title': 'Noise Reduction Method for Radio Astronomy Single Station Observation Based on Wavelet Transform and Mathematical Morphology', 'abstract': 'The 21 cm radiation of neutral hydrogen provides crucial information for\nstudying the early universe and its evolution. To advance this research,\ncountries have made significant investments in constructing large low-frequency\nradio telescope arrays, such as the Low Frequency Array (LOFAR) and the Square\nKilometre Array Phase 1 Low Frequency (SKA1-low). These instruments are pivotal\nfor radio astronomy research. However, challenges such as ionospheric plasma\ninterference, ambient radio noise, and instrument-related effects have become\nincreasingly prominent, posing major obstacles in cosmology research. To\naddress these issues, this paper proposes an efficient signal processing method\nthat combines wavelet transform and mathematical morphology. The method\ninvolves the following steps: Background Subtraction: Background interference\nin radio observation signals is eliminated. Wavelet Transform: The signal,\nafter removing background noise, undergoes a two-dimensional discrete wavelet\ntransform. Threshold processing is then applied to the wavelet coefficients to\neffectively remove interference components. Wavelet Inversion: The processed\nsignal is reconstructed using wavelet inversion. Mathematical Morphology: The\nreconstructed signal is further optimized using mathematical morphology to\nrefine the results. Experimental verification was conducted using solar\nobservation data from the Xinjiang Observatory and the Yunnan Observatory. The\nresults demonstrate that this method successfully removes interference signals\nwhile preserving useful signals, thus improving the accuracy of radio astronomy\nobservations and reducing the impact of radio frequency interference (RFI).', 'authors': ['Ming-wei Qin', 'Rui Tang', 'Ying-hui Zhou', 'Chang-jun Lan', 'Wen-hao Fu', 'Huan Wang', 'Bao-lin Hou', 'Zamri', 'Jin-song Ping', 'Wen-jun Yang', 'Liang Dong'], 'published': '2025-08-01 07:28:21+00:00', 'categories': ['astro-ph.IM', 'astro-ph.SR'], 'url': 'http://arxiv.org/abs/2508.00386v1'}
296
2508.00319
2,508.00319
null
Steering Guidance for Personalized Text-to-Image Diffusion Models
Personalizing text-to-image diffusion models is crucial for adapting the pre-trained models to specific target concepts, enabling diverse image generation. However, fine-tuning with few images introduces an inherent trade-off between aligning with the target distribution (e.g., subject fidelity) and preserving the broad knowledge of the original model (e.g., text editability). Existing sampling guidance methods, such as classifier-free guidance (CFG) and autoguidance (AG), fail to effectively guide the output toward well-balanced space: CFG restricts the adaptation to the target distribution, while AG compromises text alignment. To address these limitations, we propose personalization guidance, a simple yet effective method leveraging an unlearned weak model conditioned on a null text prompt. Moreover, our method dynamically controls the extent of unlearning in a weak model through weight interpolation between pre-trained and fine-tuned models during inference. Unlike existing guidance methods, which depend solely on guidance scales, our method explicitly steers the outputs toward a balanced latent space without additional computational overhead. Experimental results demonstrate that our proposed guidance can improve text alignment and target distribution fidelity, integrating seamlessly with various fine-tuning strategies.
2025-08-01
{'id': '2508.00319', 'title': 'Steering Guidance for Personalized Text-to-Image Diffusion Models', 'abstract': 'Personalizing text-to-image diffusion models is crucial for adapting the\npre-trained models to specific target concepts, enabling diverse image\ngeneration. However, fine-tuning with few images introduces an inherent\ntrade-off between aligning with the target distribution (e.g., subject\nfidelity) and preserving the broad knowledge of the original model (e.g., text\neditability). Existing sampling guidance methods, such as classifier-free\nguidance (CFG) and autoguidance (AG), fail to effectively guide the output\ntoward well-balanced space: CFG restricts the adaptation to the target\ndistribution, while AG compromises text alignment. To address these\nlimitations, we propose personalization guidance, a simple yet effective method\nleveraging an unlearned weak model conditioned on a null text prompt. Moreover,\nour method dynamically controls the extent of unlearning in a weak model\nthrough weight interpolation between pre-trained and fine-tuned models during\ninference. Unlike existing guidance methods, which depend solely on guidance\nscales, our method explicitly steers the outputs toward a balanced latent space\nwithout additional computational overhead. Experimental results demonstrate\nthat our proposed guidance can improve text alignment and target distribution\nfidelity, integrating seamlessly with various fine-tuning strategies.', 'authors': ['Sunghyun Park', 'Seokeon Choi', 'Hyoungwoo Park', 'Sungrack Yun'], 'published': '2025-08-01 05:02:26+00:00', 'categories': ['cs.CV', 'cs.LG'], 'url': 'http://arxiv.org/abs/2508.00319v1'}
297
2508.00591
2,508.00591
null
Wukong Framework for Not Safe For Work Detection in Text-to-Image systems
Text-to-Image (T2I) generation is a popular AI-generated content (AIGC) technology enabling diverse and creative image synthesis. However, some outputs may contain Not Safe For Work (NSFW) content (e.g., violence), violating community guidelines. Detecting NSFW content efficiently and accurately, known as external safeguarding, is essential. Existing external safeguards fall into two types: text filters, which analyze user prompts but overlook T2I model-specific variations and are prone to adversarial attacks; and image filters, which analyze final generated images but are computationally costly and introduce latency. Diffusion models, the foundation of modern T2I systems like Stable Diffusion, generate images through iterative denoising using a U-Net architecture with ResNet and Transformer blocks. We observe that: (1) early denoising steps define the semantic layout of the image, and (2) cross-attention layers in U-Net are crucial for aligning text and image regions. Based on these insights, we propose Wukong, a transformer-based NSFW detection framework that leverages intermediate outputs from early denoising steps and reuses U-Net's pre-trained cross-attention parameters. Wukong operates within the diffusion process, enabling early detection without waiting for full image generation. We also introduce a new dataset containing prompts, seeds, and image-specific NSFW labels, and evaluate Wukong on this and two public benchmarks. Results show that Wukong significantly outperforms text-based safeguards and achieves comparable accuracy of image filters, while offering much greater efficiency.
2025-08-01
{'id': '2508.00591', 'title': 'Wukong Framework for Not Safe For Work Detection in Text-to-Image systems', 'abstract': "Text-to-Image (T2I) generation is a popular AI-generated content (AIGC)\ntechnology enabling diverse and creative image synthesis. However, some outputs\nmay contain Not Safe For Work (NSFW) content (e.g., violence), violating\ncommunity guidelines. Detecting NSFW content efficiently and accurately, known\nas external safeguarding, is essential. Existing external safeguards fall into\ntwo types: text filters, which analyze user prompts but overlook T2I\nmodel-specific variations and are prone to adversarial attacks; and image\nfilters, which analyze final generated images but are computationally costly\nand introduce latency. Diffusion models, the foundation of modern T2I systems\nlike Stable Diffusion, generate images through iterative denoising using a\nU-Net architecture with ResNet and Transformer blocks. We observe that: (1)\nearly denoising steps define the semantic layout of the image, and (2)\ncross-attention layers in U-Net are crucial for aligning text and image\nregions. Based on these insights, we propose Wukong, a transformer-based NSFW\ndetection framework that leverages intermediate outputs from early denoising\nsteps and reuses U-Net's pre-trained cross-attention parameters. Wukong\noperates within the diffusion process, enabling early detection without waiting\nfor full image generation. We also introduce a new dataset containing prompts,\nseeds, and image-specific NSFW labels, and evaluate Wukong on this and two\npublic benchmarks. Results show that Wukong significantly outperforms\ntext-based safeguards and achieves comparable accuracy of image filters, while\noffering much greater efficiency.", 'authors': ['Mingrui Liu', 'Sixiao Zhang', 'Cheng Long'], 'published': '2025-08-01 12:45:30+00:00', 'categories': ['cs.CV', 'cs.AI', 'cs.CR'], 'url': 'http://arxiv.org/abs/2508.00591v1'}
298
2508.00580
2,508.0058
null
OmniUnet: A Multimodal Network for Unstructured Terrain Segmentation on Planetary Rovers Using RGB, Depth, and Thermal Imagery
Robot navigation in unstructured environments requires multimodal perception systems that can support safe navigation. Multimodality enables the integration of complementary information collected by different sensors. However, this information must be processed by machine learning algorithms specifically designed to leverage heterogeneous data. Furthermore, it is necessary to identify which sensor modalities are most informative for navigation in the target environment. In Martian exploration, thermal imagery has proven valuable for assessing terrain safety due to differences in thermal behaviour between soil types. This work presents OmniUnet, a transformer-based neural network architecture for semantic segmentation using RGB, depth, and thermal (RGB-D-T) imagery. A custom multimodal sensor housing was developed using 3D printing and mounted on the Martian Rover Testbed for Autonomy (MaRTA) to collect a multimodal dataset in the Bardenas semi-desert in northern Spain. This location serves as a representative environment of the Martian surface, featuring terrain types such as sand, bedrock, and compact soil. A subset of this dataset was manually labeled to support supervised training of the network. The model was evaluated both quantitatively and qualitatively, achieving a pixel accuracy of 80.37% and demonstrating strong performance in segmenting complex unstructured terrain. Inference tests yielded an average prediction time of 673 ms on a resource-constrained computer (Jetson Orin Nano), confirming its suitability for on-robot deployment. The software implementation of the network and the labeled dataset have been made publicly available to support future research in multimodal terrain perception for planetary robotics.
2025-08-01
{'id': '2508.00580', 'title': 'OmniUnet: A Multimodal Network for Unstructured Terrain Segmentation on Planetary Rovers Using RGB, Depth, and Thermal Imagery', 'abstract': 'Robot navigation in unstructured environments requires multimodal perception\nsystems that can support safe navigation. Multimodality enables the integration\nof complementary information collected by different sensors. However, this\ninformation must be processed by machine learning algorithms specifically\ndesigned to leverage heterogeneous data. Furthermore, it is necessary to\nidentify which sensor modalities are most informative for navigation in the\ntarget environment. In Martian exploration, thermal imagery has proven valuable\nfor assessing terrain safety due to differences in thermal behaviour between\nsoil types. This work presents OmniUnet, a transformer-based neural network\narchitecture for semantic segmentation using RGB, depth, and thermal (RGB-D-T)\nimagery. A custom multimodal sensor housing was developed using 3D printing and\nmounted on the Martian Rover Testbed for Autonomy (MaRTA) to collect a\nmultimodal dataset in the Bardenas semi-desert in northern Spain. This location\nserves as a representative environment of the Martian surface, featuring\nterrain types such as sand, bedrock, and compact soil. A subset of this dataset\nwas manually labeled to support supervised training of the network. The model\nwas evaluated both quantitatively and qualitatively, achieving a pixel accuracy\nof 80.37% and demonstrating strong performance in segmenting complex\nunstructured terrain. Inference tests yielded an average prediction time of 673\nms on a resource-constrained computer (Jetson Orin Nano), confirming its\nsuitability for on-robot deployment. The software implementation of the network\nand the labeled dataset have been made publicly available to support future\nresearch in multimodal terrain perception for planetary robotics.', 'authors': ['Raul Castilla-Arquillo', 'Carlos Perez-del-Pulgar', 'Levin Gerdes', 'Alfonso Garcia-Cerezo', 'Miguel A. Olivares-Mendez'], 'published': '2025-08-01 12:23:29+00:00', 'categories': ['cs.RO', 'cs.AI'], 'url': 'http://arxiv.org/abs/2508.00580v1'}