id
string | submitter
string | authors
string | title
string | comments
string | journal-ref
string | doi
string | report-no
string | categories
string | license
string | abstract
string | versions
list | update_date
timestamp[s] | authors_parsed
sequence |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2505.10554 | Zhiyuan Hu | Zhiyuan Hu, Yibo Wang, Hanze Dong, Yuhui Xu, Amrita Saha, Caiming Xiong, Bryan Hooi, Junnan Li | Beyond 'Aha!': Toward Systematic Meta-Abilities Alignment in Large Reasoning Models | In Progress | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large reasoning models (LRMs) already possess a latent capacity for long chain-of-thought reasoning. Prior work has shown that outcome-based reinforcement learning (RL) can incidentally elicit advanced reasoning behaviors such as self-correction, backtracking, and verification phenomena often referred to as the model's "aha moment". However, the timing and consistency of these emergent behaviors remain unpredictable and uncontrollable, limiting the scalability and reliability of LRMs' reasoning capabilities. To address these limitations, we move beyond reliance on prompts and coincidental "aha moments". Instead, we explicitly align models with three meta-abilities: deduction, induction, and abduction, using automatically generated, self-verifiable tasks. Our three stage-pipeline individual alignment, parameter-space merging, and domain-specific reinforcement learning, boosting performance by over 10\% relative to instruction-tuned baselines. Furthermore, domain-specific RL from the aligned checkpoint yields an additional gain in performance ceiling for both 7B and 32B models across math, coding, and science benchmarks, demonstrating that explicit meta-ability alignment offers a scalable and dependable foundation for reasoning. Code is available at: https://github.com/zhiyuanhubj/Meta-Ability-Alignment | [
{
"version": "v1",
"created": "Thu, 15 May 2025 17:58:33 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 17:41:22 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Hu",
"Zhiyuan",
""
],
[
"Wang",
"Yibo",
""
],
[
"Dong",
"Hanze",
""
],
[
"Xu",
"Yuhui",
""
],
[
"Saha",
"Amrita",
""
],
[
"Xiong",
"Caiming",
""
],
[
"Hooi",
"Bryan",
""
],
[
"Li",
"Junnan",
""
]
] |
2505.10610 | Zhaowei Wang | Zhaowei Wang, Wenhao Yu, Xiyu Ren, Jipeng Zhang, Yu Zhao, Rohit Saxena, Liang Cheng, Ginny Wong, Simon See, Pasquale Minervini, Yangqiu Song, and Mark Steedman | MMLongBench: Benchmarking Long-Context Vision-Language Models Effectively and Thoroughly | Work in progress | null | null | null | cs.CV cs.CL | http://creativecommons.org/publicdomain/zero/1.0/ | The rapid extension of context windows in large vision-language models has given rise to long-context vision-language models (LCVLMs), which are capable of handling hundreds of images with interleaved text tokens in a single forward pass. In this work, we introduce MMLongBench, the first benchmark covering a diverse set of long-context vision-language tasks, to evaluate LCVLMs effectively and thoroughly. MMLongBench is composed of 13,331 examples spanning five different categories of downstream tasks, such as Visual RAG and Many-Shot ICL. It also provides broad coverage of image types, including various natural and synthetic images. To assess the robustness of the models to different input lengths, all examples are delivered at five standardized input lengths (8K-128K tokens) via a cross-modal tokenization scheme that combines vision patches and text tokens. Through a thorough benchmarking of 46 closed-source and open-source LCVLMs, we provide a comprehensive analysis of the current models' vision-language long-context ability. Our results show that: i) performance on a single task is a weak proxy for overall long-context capability; ii) both closed-source and open-source models face challenges in long-context vision-language tasks, indicating substantial room for future improvement; iii) models with stronger reasoning ability tend to exhibit better long-context performance. By offering wide task coverage, various image types, and rigorous length control, MMLongBench provides the missing foundation for diagnosing and advancing the next generation of LCVLMs. | [
{
"version": "v1",
"created": "Thu, 15 May 2025 17:52:54 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 21:29:07 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wang",
"Zhaowei",
""
],
[
"Yu",
"Wenhao",
""
],
[
"Ren",
"Xiyu",
""
],
[
"Zhang",
"Jipeng",
""
],
[
"Zhao",
"Yu",
""
],
[
"Saxena",
"Rohit",
""
],
[
"Cheng",
"Liang",
""
],
[
"Wong",
"Ginny",
""
],
[
"See",
"Simon",
""
],
[
"Minervini",
"Pasquale",
""
],
[
"Song",
"Yangqiu",
""
],
[
"Steedman",
"Mark",
""
]
] |
2505.10615 | Muldrow Etheredge | Muldrow Etheredge | Taxonomy of branes in infinite distance limits | 119 pages, 33 figures | null | null | ACFI-T25-02 | hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | I consider flat slices of moduli spaces where the $(-\nabla \log T)$-vectors of particle-towers and branes are constant, and I show that the Emergent String Conjecture constrains these vectors to reside on lattices. In asymptotic limits, this results in exponentially separated, discretized hierarchies of energy scales. I further identify conditions that determine whether a given lattice site must be populated, and I show that only a finite set of configurations satisfies these conditions. I classify all such configurations for 0d, 1d, and 2d moduli spaces in theories with 3 to 11 spacetime dimensions, and I argue that 11d is the maximal spacetime dimension compatible with my assumptions. Remarkably, this classification reproduces the detailed particle and brane content of various string theory examples with 32, 16, and 8 supercharges. It also describes some examples where the assumptions I use are violated, suggesting that my assumptions can be relaxed and the scope of this classification can be expanded. It might also predict new branes. For instance, if heterotic string theory is described by this classification, then it must possess non-BPS branes with D-brane-like tensions. Similarly, if this classification applies to the Dark Dimension Scenario with an extra modulus, then it requires the existence of strings with tensions related to the cosmological constant by $T\lesssim \Lambda^{1/6}$ in 4d Planck units. | [
{
"version": "v1",
"created": "Thu, 15 May 2025 18:00:00 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 17:05:58 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Etheredge",
"Muldrow",
""
]
] |
2505.10879 | Ali Sartaz Khan | Ali Sartaz Khan, Tolulope Ogunremi, Ahmed Adel Attia, Dorottya Demszky | Multi-Stage Speaker Diarization for Noisy Classrooms | null | null | null | null | cs.SD cs.LG eess.AS | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Speaker diarization, the process of identifying "who spoke when" in audio recordings, is essential for understanding classroom dynamics. However, classroom settings present distinct challenges, including poor recording quality, high levels of background noise, overlapping speech, and the difficulty of accurately capturing children's voices. This study investigates the effectiveness of multi-stage diarization models using Nvidia's NeMo diarization pipeline. We assess the impact of denoising on diarization accuracy and compare various voice activity detection (VAD) models, including self-supervised transformer-based frame-wise VAD models. We also explore a hybrid VAD approach that integrates Automatic Speech Recognition (ASR) word-level timestamps with frame-level VAD predictions. We conduct experiments using two datasets from English speaking classrooms to separate teacher vs. student speech and to separate all speakers. Our results show that denoising significantly improves the Diarization Error Rate (DER) by reducing the rate of missed speech. Additionally, training on both denoised and noisy datasets leads to substantial performance gains in noisy conditions. The hybrid VAD model leads to further improvements in speech detection, achieving a DER as low as 17% in teacher-student experiments and 45% in all-speaker experiments. However, we also identified trade-offs between voice activity detection and speaker confusion. Overall, our study highlights the effectiveness of multi-stage diarization models and integrating ASR-based information for enhancing speaker diarization in noisy classroom environments. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 05:35:06 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 06:02:39 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Khan",
"Ali Sartaz",
""
],
[
"Ogunremi",
"Tolulope",
""
],
[
"Attia",
"Ahmed Adel",
""
],
[
"Demszky",
"Dorottya",
""
]
] |
2505.10926 | Wataru Chiba | Wataru Chiba, Ryusuke Jinno, Kimihiro Nomura | Graviton-photon conversion in stochastic magnetic fields | 29 pages, 6 figures | null | null | KOBE-COSMO-25-08, KUNS-3048 | gr-qc astro-ph.CO hep-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We study graviton-photon conversion in the presence of stochastic magnetic fields. Assuming Gaussian magnetic fields that may possess nontrivial helicity, and unpolarized gravitational waves (GWs) as the initial state, we obtain expressions for the intensity and linear/circular polarizations of GWs after propagation over a finite distance. We calculate both the expectation values and variances of these observables, and find their nontrivial dependence on the typical correlation length of the magnetic field, the propagation distance, and the photon plasma mass. Our analysis reveals that an observationally favorable frequency range with narrower variance can emerge for the intensity, while a peak structure appears in the expectation value of the circular polarization when the magnetic field has nonzero helicity. We also identify a consistency relation between the GW intensity and circular polarization. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 07:01:22 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 10:30:18 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Chiba",
"Wataru",
""
],
[
"Jinno",
"Ryusuke",
""
],
[
"Nomura",
"Kimihiro",
""
]
] |
2505.11015 | Jingqun Tang | An-Lan Wang, Jingqun Tang, Liao Lei, Hao Feng, Qi Liu, Xiang Fei, Jinghui Lu, Han Wang, Weiwei Liu, Hao Liu, Yuliang Liu, Xiang Bai, Can Huang | WildDoc: How Far Are We from Achieving Comprehensive and Robust Document Understanding in the Wild? | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by-nc-sa/4.0/ | The rapid advancements in Multimodal Large Language Models (MLLMs) have significantly enhanced capabilities in Document Understanding. However, prevailing benchmarks like DocVQA and ChartQA predominantly comprise \textit{scanned or digital} documents, inadequately reflecting the intricate challenges posed by diverse real-world scenarios, such as variable illumination and physical distortions. This paper introduces WildDoc, the inaugural benchmark designed specifically for assessing document understanding in natural environments. WildDoc incorporates a diverse set of manually captured document images reflecting real-world conditions and leverages document sources from established benchmarks to facilitate comprehensive comparisons with digital or scanned documents. Further, to rigorously evaluate model robustness, each document is captured four times under different conditions. Evaluations of state-of-the-art MLLMs on WildDoc expose substantial performance declines and underscore the models' inadequate robustness compared to traditional benchmarks, highlighting the unique challenges posed by real-world document understanding. Our project homepage is available at https://bytedance.github.io/WildDoc. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 09:09:46 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:00:24 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wang",
"An-Lan",
""
],
[
"Tang",
"Jingqun",
""
],
[
"Lei",
"Liao",
""
],
[
"Feng",
"Hao",
""
],
[
"Liu",
"Qi",
""
],
[
"Fei",
"Xiang",
""
],
[
"Lu",
"Jinghui",
""
],
[
"Wang",
"Han",
""
],
[
"Liu",
"Weiwei",
""
],
[
"Liu",
"Hao",
""
],
[
"Liu",
"Yuliang",
""
],
[
"Bai",
"Xiang",
""
],
[
"Huang",
"Can",
""
]
] |
2505.11048 | Ana Mitrasinovic | Ana Mitra\v{s}inovi\'c, Branislav Vukoti\'c, Teodora \v{Z}i\v{z}ak, Miroslav Micic, Milan M. \'Cirkovi\'c | Revisiting the bimodality of galactic habitability in IllustrisTNG | Accepted for publication in PASA | null | 10.1017/pasa.2025.10041 | null | astro-ph.GA astro-ph.EP | http://creativecommons.org/licenses/by-sa/4.0/ | The potential of galaxies to host habitable planets is central to astrobiology, tightly linked to galaxy-scale evolution and cosmological processes. Using IllustrisTNG, we revisit the proposed local peak in the mass-metallicity relation for small, metal-rich, star-forming galaxies (Cloudlet) as an indicator of enhanced galactic habitability. We refine the earlier analysis by applying updated filtering criteria to identify a more refined sample, further selecting objects based on their history. This process resulted in a confirmed sample of 97 dwarf galaxies, alongside 519 additional structures of uncertain origin, potentially comprising both numerical artefacts and unrecognised physical systems. Under these stricter conditions, the proposed bimodality in galactic habitability is strongly diminished. However, the astrobiological potential of metal-rich dwarfs, most of which are compact remnants of more massive galaxies that underwent tidal stripping, is a thrilling area of exploration. Although dense stellar environments are traditionally seen as inhospitable, recent studies highlight the role of dynamic environments in enhancing the distribution of biological material. Furthermore, the potential habitability of tidal structures formed in the aftermath of galactic interactions is a fascinating possibility. Our findings suggest that non-traditional structures support conditions favourable for life, opening up exciting new avenues for astrobiological research. This research underscores the need for a holistic approach to studying habitability that moves beyond planetary and stellar-focused frameworks to incorporate the broader galactic environment. Understanding the interactions between galaxies, their evolution, and the influence of their surroundings is essential to developing a more comprehensive model of how and where life might emerge and persist across the Universe. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 09:42:50 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Mitrašinović",
"Ana",
""
],
[
"Vukotić",
"Branislav",
""
],
[
"Žižak",
"Teodora",
""
],
[
"Micic",
"Miroslav",
""
],
[
"Ćirković",
"Milan M.",
""
]
] |
2505.11158 | Xiangcheng Liu | Xing Hu, Xiangcheng Liu, Danfeng Hong, Qianqian Duan, Linghua Jiang, Haima Yang and Dawei Zhan | Recent Advances in Diffusion Models for Hyperspectral Image Processing and Analysis: A Review | null | null | null | null | eess.IV cs.CV | http://creativecommons.org/licenses/by-sa/4.0/ | Hyperspectral image processing and analysis has important application value in remote sensing, agriculture and environmental monitoring, but its high dimensionality, data redundancy and noise interference etc. bring great challenges to the analysis. Traditional models have limitations in dealing with these complex data, and it is difficult to meet the increasing demand for analysis. In recent years, Diffusion models, as a class of emerging generative approaches, have demonstrated promising capabilities in hyperspectral image (HSI) processing tasks. By simulating the diffusion process of data in time, the Diffusion Model are capable of modeling high-dimensional spectral structures, generate high-quality samples, and achieve competitive performance in spectral-spatial denoising tasks and data enhancement. In this paper, we review the recent research advances in diffusion modeling for hyperspectral image processing and analysis, and discuss its applications in tasks such as high-dimensional data processing, noise removal, classification, and anomaly detection. The performance of diffusion-based models on image processing is compared and the challenges are summarized. It is shown that the diffusion model can significantly improve the accuracy and efficiency of hyperspectral image analysis, providing a new direction for future research. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 11:59:48 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:52:03 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Hu",
"Xing",
""
],
[
"Liu",
"Xiangcheng",
""
],
[
"Hong",
"Danfeng",
""
],
[
"Duan",
"Qianqian",
""
],
[
"Jiang",
"Linghua",
""
],
[
"Yang",
"Haima",
""
],
[
"Zhan",
"Dawei",
""
]
] |
2505.11210 | Anders Gj{\o}lbye | Anders Gj{\o}lbye, Stefan Haufe, Lars Kai Hansen | Minimizing False-Positive Attributions in Explanations of Non-Linear Models | Preprint. Under review | null | null | null | cs.LG stat.ML | http://creativecommons.org/licenses/by/4.0/ | Suppressor variables can influence model predictions without being dependent on the target outcome and they pose a significant challenge for Explainable AI (XAI) methods. These variables may cause false-positive feature attributions, undermining the utility of explanations. Although effective remedies exist for linear models, their extension to non-linear models and to instance-based explanations has remained limited. We introduce PatternLocal, a novel XAI technique that addresses this gap. PatternLocal begins with a locally linear surrogate, e.g. LIME, KernelSHAP, or gradient-based methods, and transforms the resulting discriminative model weights into a generative representation, thereby suppressing the influence of suppressor variables while preserving local fidelity. In extensive hyperparameter optimization on the XAI-TRIS benchmark, PatternLocal consistently outperformed other XAI methods and reduced false-positive attributions when explaining non-linear tasks, thereby enabling more reliable and actionable insights. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 13:06:12 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:24:58 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Gjølbye",
"Anders",
""
],
[
"Haufe",
"Stefan",
""
],
[
"Hansen",
"Lars Kai",
""
]
] |
2505.11312 | Emanuele Francazi | Emanuele Francazi, Francesco Pinto, Aurelien Lucchi, Marco Baity-Jesi | Where You Place the Norm Matters: From Prejudiced to Neutral Initializations | This version includes minor revisions. These changes do not affect the main results or conclusions | null | null | null | cs.LG cond-mat.dis-nn | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Normalization layers, such as Batch Normalization and Layer Normalization, are central components in modern neural networks, widely adopted to improve training stability and generalization. While their practical effectiveness is well documented, a detailed theoretical understanding of how normalization affects model behavior, starting from initialization, remains an important open question. In this work, we investigate how both the presence and placement of normalization within hidden layers influence the statistical properties of network predictions before training begins. In particular, we study how these choices shape the distribution of class predictions at initialization, which can range from unbiased (Neutral) to highly concentrated (Prejudiced) toward a subset of classes. Our analysis shows that normalization placement induces systematic differences in the initial prediction behavior of neural networks, which in turn shape the dynamics of learning. By linking architectural choices to prediction statistics at initialization, our work provides a principled understanding of how normalization can influence early training behavior and offers guidance for more controlled and interpretable network design. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 14:38:30 GMT"
},
{
"version": "v2",
"created": "Fri, 23 May 2025 13:12:15 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 06:51:38 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Francazi",
"Emanuele",
""
],
[
"Pinto",
"Francesco",
""
],
[
"Lucchi",
"Aurelien",
""
],
[
"Baity-Jesi",
"Marco",
""
]
] |
2505.11484 | Yige Xu | Yige Xu, Xu Guo, Zhiwei Zeng, Chunyan Miao | SoftCoT++: Test-Time Scaling with Soft Chain-of-Thought Reasoning | 14 pages | null | null | null | cs.CL | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Test-Time Scaling (TTS) refers to approaches that improve reasoning performance by allocating extra computation during inference, without altering the model's parameters. While existing TTS methods operate in a discrete token space by generating more intermediate steps, recent studies in Coconut and SoftCoT have demonstrated that thinking in the continuous latent space can further enhance the reasoning performance. Such latent thoughts encode informative thinking without the information loss associated with autoregressive token generation, sparking increased interest in continuous-space reasoning. Unlike discrete decoding, where repeated sampling enables exploring diverse reasoning paths, latent representations in continuous space are fixed for a given input, which limits diverse exploration, as all decoded paths originate from the same latent thought. To overcome this limitation, we introduce SoftCoT++ to extend SoftCoT to the Test-Time Scaling paradigm by enabling diverse exploration of thinking paths. Specifically, we perturb latent thoughts via multiple specialized initial tokens and apply contrastive learning to promote diversity among soft thought representations. Experiments across five reasoning benchmarks and two distinct LLM architectures demonstrate that SoftCoT++ significantly boosts SoftCoT and also outperforms SoftCoT with self-consistency scaling. Moreover, it shows strong compatibility with conventional scaling techniques such as self-consistency. Source code is available at https://github.com/xuyige/SoftCoT. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 17:47:50 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:35:29 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Xu",
"Yige",
""
],
[
"Guo",
"Xu",
""
],
[
"Zeng",
"Zhiwei",
""
],
[
"Miao",
"Chunyan",
""
]
] |
2505.11618 | Pengrui Quan | Pengrui Quan, Brian Wang, Kang Yang, Liying Han, Mani Srivastava | Benchmarking Spatiotemporal Reasoning in LLMs and Reasoning Models: Capabilities and Challenges | null | null | null | null | cs.AI cs.LG eess.SP | http://creativecommons.org/licenses/by/4.0/ | Spatiotemporal reasoning plays a key role in Cyber-Physical Systems (CPS). Despite advances in Large Language Models (LLMs) and Large Reasoning Models (LRMs), their capacity to reason about complex spatiotemporal signals remains underexplored. This paper proposes a hierarchical SpatioTemporal reAsoning benchmaRK, STARK, to systematically evaluate LLMs across three levels of reasoning complexity: state estimation (e.g., predicting field variables, localizing and tracking events in space and time), spatiotemporal reasoning over states (e.g., inferring spatial-temporal relationships), and world-knowledge-aware reasoning that integrates contextual and domain knowledge (e.g., intent prediction, landmark-aware navigation). We curate 26 distinct spatiotemporal tasks with diverse sensor modalities, comprising 14,552 challenges where models answer directly or by Python Code Interpreter. Evaluating 3 LRMs and 8 LLMs, we find LLMs achieve limited success in tasks requiring geometric reasoning (e.g., multilateration or triangulation), particularly as complexity increases. Surprisingly, LRMs show robust performance across tasks with various levels of difficulty, often competing or surpassing traditional first-principle-based methods. Our results show that in reasoning tasks requiring world knowledge, the performance gap between LLMs and LRMs narrows, with some LLMs even surpassing LRMs. However, the LRM o3 model continues to achieve leading performance across all evaluated tasks, a result attributed primarily to the larger size of the reasoning models. STARK motivates future innovations in model architectures and reasoning paradigms for intelligent CPS by providing a structured framework to identify limitations in the spatiotemporal reasoning of LLMs and LRMs. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 18:32:35 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 16:52:19 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Quan",
"Pengrui",
""
],
[
"Wang",
"Brian",
""
],
[
"Yang",
"Kang",
""
],
[
"Han",
"Liying",
""
],
[
"Srivastava",
"Mani",
""
]
] |
2505.11642 | Xi Li | Falong Fan and Xi Li | PeerGuard: Defending Multi-Agent Systems Against Backdoor Attacks Through Mutual Reasoning | This paper has been accepted to IEEE IRI 2025 | null | null | null | cs.MA cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Multi-agent systems leverage advanced AI models as autonomous agents that interact, cooperate, or compete to complete complex tasks across applications such as robotics and traffic management. Despite their growing importance, safety in multi-agent systems remains largely underexplored, with most research focusing on single AI models rather than interacting agents. This work investigates backdoor vulnerabilities in multi-agent systems and proposes a defense mechanism based on agent interactions. By leveraging reasoning abilities, each agent evaluates responses from others to detect illogical reasoning processes, which indicate poisoned agents. Experiments on LLM-based multi-agent systems, including ChatGPT series and Llama 3, demonstrate the effectiveness of the proposed method, achieving high accuracy in identifying poisoned agents while minimizing false positives on clean agents. We believe this work provides insights into multi-agent system safety and contributes to the development of robust, trustworthy AI interactions. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 19:08:29 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 15:31:58 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Fan",
"Falong",
""
],
[
"Li",
"Xi",
""
]
] |
2505.11696 | Beatriz Fern\'andez-Dom\'inguez | J. Lois-Fuentes, B. Fern\'andez-Dom\'inguez, F. Delaunay, X. Pereira-L\'opez, N.A. Orr, M. P{\l}oszajczak, N. Michel, T. Otsuka, T. Suzuki, W.N. Catford, O. Sorlin, N.L. Achouri, M. Assi\'e, S. Bailey, B. Bastin, Y. Blumenfeld, R. Borcea, M. Caama\~no, L. Caceres, E. Cl\'ement, A. Corsi, N. Curtis, Q. Deshayes, F. Farget, M. Fisichella, G. de France, S. Franchoo, M. Freer, J. Gibelin, A. Gillibert, G.F. Grinyer, F. Hammache, O. Kamalou, A. Knapton, Tz. Kokalova, V. Lapoux, J.A. Lay, B. Le Crom, S. Leblond, F.M. Marqu\'es, A. Matta, P. Morfouace, A. Moro, J. Pancin, L. Perrot, J. Piot, E. Pollacco, P. Punta, D. Ramos, C. Rodr\'iguez-Tajes, T. Roger, F. Rotaru, M. S\'enoville, N. de S\'er\'eville, R. Smith, M. Stanoiu, I. Stefani, C. Stodel, D. Suzuki, J.C. Thomas, N. Timofeyuk, M. Vandebrouck, J. Walshe, C. Wheldon | Unbound neutron $\nu0d_{3/2}$ strength in $^{17}$C and the N=16 shell gap | null | null | null | null | nucl-ex nucl-th | http://creativecommons.org/licenses/by/4.0/ | Significant continuum strength has been observed to be populated in $^{17}$C produced in the d($^{16}$C,p) reaction at a beam energy of 17.2~MeV/nucleon. The strength appears at greater than $\sim$2~MeV above the single-neutron decay threshold and has been identified as arising from transfer into the neutron $0d_{3/2}$ orbital. Guided by shell model predictions the greater majority of the strength is associated with a 3/2$^+$ state at an excitation energy of 4.40$_{-0.14}^{+0.33}$ MeV and a much weaker 3/2$^+$ level at 5.60$_{-0.45}^{+1.35}$ MeV. The corresponding total widths were determined to be 3.45$_{-0.78}^{+1.82}$ and 1.6$_{-1.4}^{+4.6}$ MeV, respectively. From the backward angle proton differential cross sections and the branching ratios for neutron decay to the $^{16}$C(2$_{1}^{+}$) level, the corresponding spectroscopic factors to the ground state were deduced to be 0.47$\pm{10}$ and $<$0.09. Shell-model calculations employing the phenomenological SFO-tls interaction as well as Gamow Shell-Model calculations including continuum effects are in reasonable agreement with experiment, although the predicted strength lies at somewhat lower energy. The size of the N=16 shell gap ($\varepsilon_{ \nu0d_{3/2}}-\varepsilon _{\nu 1s_{1/2}}$) was estimated to be 5.08$_{-0.33}^{+0.43}$~MeV - some 1.3~MeV larger than found in the SFO-tls shell model calculation. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 21:04:26 GMT"
},
{
"version": "v2",
"created": "Tue, 20 May 2025 18:24:00 GMT"
},
{
"version": "v3",
"created": "Mon, 26 May 2025 22:22:32 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Lois-Fuentes",
"J.",
""
],
[
"Fernández-Domínguez",
"B.",
""
],
[
"Delaunay",
"F.",
""
],
[
"Pereira-López",
"X.",
""
],
[
"Orr",
"N. A.",
""
],
[
"Płoszajczak",
"M.",
""
],
[
"Michel",
"N.",
""
],
[
"Otsuka",
"T.",
""
],
[
"Suzuki",
"T.",
""
],
[
"Catford",
"W. N.",
""
],
[
"Sorlin",
"O.",
""
],
[
"Achouri",
"N. L.",
""
],
[
"Assié",
"M.",
""
],
[
"Bailey",
"S.",
""
],
[
"Bastin",
"B.",
""
],
[
"Blumenfeld",
"Y.",
""
],
[
"Borcea",
"R.",
""
],
[
"Caamaño",
"M.",
""
],
[
"Caceres",
"L.",
""
],
[
"Clément",
"E.",
""
],
[
"Corsi",
"A.",
""
],
[
"Curtis",
"N.",
""
],
[
"Deshayes",
"Q.",
""
],
[
"Farget",
"F.",
""
],
[
"Fisichella",
"M.",
""
],
[
"de France",
"G.",
""
],
[
"Franchoo",
"S.",
""
],
[
"Freer",
"M.",
""
],
[
"Gibelin",
"J.",
""
],
[
"Gillibert",
"A.",
""
],
[
"Grinyer",
"G. F.",
""
],
[
"Hammache",
"F.",
""
],
[
"Kamalou",
"O.",
""
],
[
"Knapton",
"A.",
""
],
[
"Kokalova",
"Tz.",
""
],
[
"Lapoux",
"V.",
""
],
[
"Lay",
"J. A.",
""
],
[
"Crom",
"B. Le",
""
],
[
"Leblond",
"S.",
""
],
[
"Marqués",
"F. M.",
""
],
[
"Matta",
"A.",
""
],
[
"Morfouace",
"P.",
""
],
[
"Moro",
"A.",
""
],
[
"Pancin",
"J.",
""
],
[
"Perrot",
"L.",
""
],
[
"Piot",
"J.",
""
],
[
"Pollacco",
"E.",
""
],
[
"Punta",
"P.",
""
],
[
"Ramos",
"D.",
""
],
[
"Rodríguez-Tajes",
"C.",
""
],
[
"Roger",
"T.",
""
],
[
"Rotaru",
"F.",
""
],
[
"Sénoville",
"M.",
""
],
[
"de Séréville",
"N.",
""
],
[
"Smith",
"R.",
""
],
[
"Stanoiu",
"M.",
""
],
[
"Stefani",
"I.",
""
],
[
"Stodel",
"C.",
""
],
[
"Suzuki",
"D.",
""
],
[
"Thomas",
"J. C.",
""
],
[
"Timofeyuk",
"N.",
""
],
[
"Vandebrouck",
"M.",
""
],
[
"Walshe",
"J.",
""
],
[
"Wheldon",
"C.",
""
]
] |
2505.11807 | Yufei Xiang | Yufei Xiang and Yiqun Shen and Yeqin Zhang and Cam-Tu Nguyen | Retrospex: Language Agent Meets Offline Reinforcement Learning Critic | 17 pages, Published in EMNLP 2024 (Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing) | Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 4650-4666, ACL Anthology, 2024 | 10.18653/v1/2024.emnlp-main.268 | null | cs.CL cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large Language Models (LLMs) possess extensive knowledge and commonsense reasoning capabilities, making them valuable for creating powerful agents. However, existing LLM agent frameworks have not fully utilized past experiences for improvement. This work introduces a new LLM-based agent framework called Retrospex, which addresses this challenge by analyzing past experiences in depth. Unlike previous approaches, Retrospex does not directly integrate experiences into the LLM's context. Instead, it combines the LLM's action likelihood with action values estimated by a Reinforcement Learning (RL) Critic, which is trained on past experiences through an offline ''retrospection'' process. Additionally, Retrospex employs a dynamic action rescoring mechanism that increases the importance of experience-based values for tasks that require more interaction with the environment. We evaluate Retrospex in ScienceWorld, ALFWorld and Webshop environments, demonstrating its advantages over strong, contemporary baselines. | [
{
"version": "v1",
"created": "Sat, 17 May 2025 03:28:24 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 01:30:17 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Xiang",
"Yufei",
""
],
[
"Shen",
"Yiqun",
""
],
[
"Zhang",
"Yeqin",
""
],
[
"Nguyen",
"Cam-Tu",
""
]
] |
2505.11808 | Hochul Hwang | Shangqun Yu, Hochul Hwang, Trung M. Dang, Joydeep Biswas, Nicholas A. Giudice, Sunghoon Ivan Lee, Donghyun Kim | Human-Centered Development of Guide Dog Robots: Quiet and Stable Locomotion Control | null | null | null | null | cs.RO cs.HC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | A quadruped robot is a promising system that can offer assistance comparable to that of dog guides due to its similar form factor. However, various challenges remain in making these robots a reliable option for blind and low-vision (BLV) individuals. Among these challenges, noise and jerky motion during walking are critical drawbacks of existing quadruped robots. While these issues have largely been overlooked in guide dog robot research, our interviews with guide dog handlers and trainers revealed that acoustic and physical disturbances can be particularly disruptive for BLV individuals, who rely heavily on environmental sounds for navigation. To address these issues, we developed a novel walking controller for slow stepping and smooth foot swing/contact while maintaining human walking speed, as well as robust and stable balance control. The controller integrates with a perception system to facilitate locomotion over non-flat terrains, such as stairs. Our controller was extensively tested on the Unitree Go1 robot and, when compared with other control methods, demonstrated significant noise reduction -- half of the default locomotion controller. In this study, we adopt a mixed-methods approach to evaluate its usability with BLV individuals. In our indoor walking experiments, participants compared our controller to the robot's default controller. Results demonstrated superior acceptance of our controller, highlighting its potential to improve the user experience of guide dog robots. Video demonstration (best viewed with audio) available at: https://youtu.be/8-pz_8Hqe6s. | [
{
"version": "v1",
"created": "Sat, 17 May 2025 03:35:58 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 12:30:37 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Yu",
"Shangqun",
""
],
[
"Hwang",
"Hochul",
""
],
[
"Dang",
"Trung M.",
""
],
[
"Biswas",
"Joydeep",
""
],
[
"Giudice",
"Nicholas A.",
""
],
[
"Lee",
"Sunghoon Ivan",
""
],
[
"Kim",
"Donghyun",
""
]
] |
2505.11998 | Prashant Bhat Mr | Prashant Shivaram Bhat, Shakib Yazdani, Elahe Arani, Bahram Zonooz | Parameter Efficient Continual Learning with Dynamic Low-Rank Adaptation | 27 pages, 5 figures | null | null | null | cs.LG cs.CV | http://creativecommons.org/licenses/by/4.0/ | Catastrophic forgetting has remained a critical challenge for deep neural networks in Continual Learning (CL) as it undermines consolidated knowledge when learning new tasks. Parameter efficient fine tuning CL techniques are gaining traction for their effectiveness in addressing catastrophic forgetting with a lightweight training schedule while avoiding degradation of consolidated knowledge in pre-trained models. However, low rank adapters (LoRA) in these approaches are highly sensitive to rank selection which can lead to sub-optimal resource allocation and performance. To this end, we introduce PEARL, a rehearsal-free CL framework that entails dynamic rank allocation for LoRA components during CL training. Specifically, PEARL leverages reference task weights and adaptively determines the rank of task-specific LoRA components based on the current tasks' proximity to reference task weights in parameter space. To demonstrate the versatility of PEARL, we evaluate it across three vision architectures (ResNet, Separable Convolutional Network and Vision Transformer) and a multitude of CL scenarios, and show that PEARL outperforms all considered baselines by a large margin. | [
{
"version": "v1",
"created": "Sat, 17 May 2025 13:19:01 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 13:19:59 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Bhat",
"Prashant Shivaram",
""
],
[
"Yazdani",
"Shakib",
""
],
[
"Arani",
"Elahe",
""
],
[
"Zonooz",
"Bahram",
""
]
] |
2505.12074 | Yuhao Yi | Chen Shu, Boyu Fu, Yiman Li, Ting Yin, Wenchuan Zhang, Jie Chen, Yuhao Yi, Hong Bu | Denoising Mutual Knowledge Distillation in Bi-Directional Multiple Instance Learning | 15 pages, 3 figures | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Multiple Instance Learning is the predominant method for Whole Slide Image classification in digital pathology, enabling the use of slide-level labels to supervise model training. Although MIL eliminates the tedious fine-grained annotation process for supervised learning, whether it can learn accurate bag- and instance-level classifiers remains a question. To address the issue, instance-level classifiers and instance masks were incorporated to ground the prediction on supporting patches. These methods, while practically improving the performance of MIL methods, may potentially introduce noisy labels. We propose to bridge the gap between commonly used MIL and fully supervised learning by augmenting both the bag- and instance-level learning processes with pseudo-label correction capabilities elicited from weak to strong generalization techniques. The proposed algorithm improves the performance of dual-level MIL algorithms on both bag- and instance-level predictions. Experiments on public pathology datasets showcase the advantage of the proposed methods. | [
{
"version": "v1",
"created": "Sat, 17 May 2025 16:26:43 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 11:14:27 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Shu",
"Chen",
""
],
[
"Fu",
"Boyu",
""
],
[
"Li",
"Yiman",
""
],
[
"Yin",
"Ting",
""
],
[
"Zhang",
"Wenchuan",
""
],
[
"Chen",
"Jie",
""
],
[
"Yi",
"Yuhao",
""
],
[
"Bu",
"Hong",
""
]
] |
2505.12155 | Ranit Karmakar | Ranit Karmakar, Simon F. N{\o}rrelykke | SoftPQ: Robust Instance Segmentation Evaluation via Soft Matching and Tunable Thresholds | null | null | null | null | cs.CV cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Segmentation evaluation metrics traditionally rely on binary decision logic: predictions are either correct or incorrect, based on rigid IoU thresholds. Detection--based metrics such as F1 and mAP determine correctness at the object level using fixed overlap cutoffs, while overlap--based metrics like Intersection over Union (IoU) and Dice operate at the pixel level, often overlooking instance--level structure. Panoptic Quality (PQ) attempts to unify detection and segmentation assessment, but it remains dependent on hard-threshold matching--treating predictions below the threshold as entirely incorrect. This binary framing obscures important distinctions between qualitatively different errors and fails to reward gradual model improvements. We propose SoftPQ, a flexible and interpretable instance segmentation metric that redefines evaluation as a graded continuum rather than a binary classification. SoftPQ introduces tunable upper and lower IoU thresholds to define a partial matching region and applies a sublinear penalty function to ambiguous or fragmented predictions. These extensions allow SoftPQ to exhibit smoother score behavior, greater robustness to structural segmentation errors, and more informative feedback for model development and evaluation. Through controlled perturbation experiments, we show that SoftPQ captures meaningful differences in segmentation quality that existing metrics overlook, making it a practical and principled alternative for both benchmarking and iterative model refinement. | [
{
"version": "v1",
"created": "Sat, 17 May 2025 22:08:33 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 01:54:05 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Karmakar",
"Ranit",
""
],
[
"Nørrelykke",
"Simon F.",
""
]
] |
2505.12180 | Bixiang Wang | Bixiang Wang | Martingale Solutions of Fractional Stochastic Reaction-Diffusion Equations Driven by Superlinear Noise | null | null | null | null | math.PR math.AP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper, we prove the existence of martingale solutions of a class of stochastic equations with pseudo-monotone drift of polynomial growth of arbitrary order and a continuous diffusion term with superlinear growth. Both the nonlinear drift and diffusion terms are not required to be locally Lipschitz continuous. We then apply the abstract result to establish the existence of martingale solutions of the fractional stochastic reaction-diffusion equation with polynomial drift driven by a superlinear noise. The pseudo-monotonicity techniques and the Skorokhod-Jakubowski representation theorem in a topological space are used to pass to the limit of a sequence of approximate solutions defined by the Galerkin method. | [
{
"version": "v1",
"created": "Sun, 18 May 2025 00:25:37 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wang",
"Bixiang",
""
]
] |
2505.12299 | Kun Huang | Kun Huang, Weikai Xu, Yuxuan Liu, Quandong Wang, Pengzhi Gao, Wei Liu, Jian Luan, Bin Wang, Bo An | Enhance Mobile Agents Thinking Process Via Iterative Preference Learning | 9 pages, 8 figures, 7 tables | null | null | null | cs.CL cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The Chain of Action-Planning Thoughts (CoaT) paradigm has been shown to improve the reasoning performance of VLM-based mobile agents in GUI tasks. However, the scarcity of diverse CoaT trajectories limits the expressiveness and generalization ability of such agents. While self-training is commonly employed to address data scarcity, existing approaches either overlook the correctness of intermediate reasoning steps or depend on expensive process-level annotations to construct process reward models (PRM). To address the above problems, we propose an Iterative Preference Learning (IPL) that constructs a CoaT-tree through interative sampling, scores leaf nodes using rule-based reward, and backpropagates feedback to derive Thinking-level Direct Preference Optimization (T-DPO) pairs. To prevent overfitting during warm-up supervised fine-tuning, we further introduce a three-stage instruction evolution, which leverages GPT-4o to generate diverse Q\&A pairs based on real mobile UI screenshots, enhancing both generality and layout understanding. Experiments on three standard Mobile GUI-agent benchmarks demonstrate that our agent MobileIPL outperforms strong baselines, including continual pretraining models such as OS-ATLAS and UI-TARS. It achieves state-of-the-art performance across three standard Mobile GUI-Agents benchmarks and shows strong generalization to out-of-domain scenarios. | [
{
"version": "v1",
"created": "Sun, 18 May 2025 08:28:05 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:49:23 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Huang",
"Kun",
""
],
[
"Xu",
"Weikai",
""
],
[
"Liu",
"Yuxuan",
""
],
[
"Wang",
"Quandong",
""
],
[
"Gao",
"Pengzhi",
""
],
[
"Liu",
"Wei",
""
],
[
"Luan",
"Jian",
""
],
[
"Wang",
"Bin",
""
],
[
"An",
"Bo",
""
]
] |
2505.12499 | Jian Xiao | Jian Xiao, Zijie Song, Jialong Hu, Hao Cheng, Zhenzhen Hu, Jia Li, Richang Hong | Rebalancing Contrastive Alignment with Learnable Semantic Gaps in Text-Video Retrieval | null | null | null | null | cs.CV cs.IR cs.MM | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent advances in text-video retrieval have been largely driven by contrastive learning frameworks. However, existing methods overlook a key source of optimization tension: the separation between text and video distributions in the representation space (referred to as the modality gap), and the prevalence of false negatives in batch sampling. These factors lead to conflicting gradients under the InfoNCE loss, impeding stable alignment. To mitigate this, we propose GARE, a Gap-Aware Retrieval framework that introduces a learnable, pair-specific increment Delta_ij between text t_i and video v_j to offload the tension from the global anchor representation. We first derive the ideal form of Delta_ij via a coupled multivariate first-order Taylor approximation of the InfoNCE loss under a trust-region constraint, revealing it as a mechanism for resolving gradient conflicts by guiding updates along a locally optimal descent direction. Due to the high cost of directly computing Delta_ij, we introduce a lightweight neural module conditioned on the semantic gap between each video-text pair, enabling structure-aware correction guided by gradient supervision. To further stabilize learning and promote interpretability, we regularize Delta using three components: a trust-region constraint to prevent oscillation, a directional diversity term to promote semantic coverage, and an information bottleneck to limit redundancy. Experiments across four retrieval benchmarks show that GARE consistently improves alignment accuracy and robustness to noisy supervision, confirming the effectiveness of gap-aware tension mitigation. | [
{
"version": "v1",
"created": "Sun, 18 May 2025 17:18:06 GMT"
},
{
"version": "v2",
"created": "Tue, 20 May 2025 07:25:42 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 02:33:49 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Xiao",
"Jian",
""
],
[
"Song",
"Zijie",
""
],
[
"Hu",
"Jialong",
""
],
[
"Cheng",
"Hao",
""
],
[
"Hu",
"Zhenzhen",
""
],
[
"Li",
"Jia",
""
],
[
"Hong",
"Richang",
""
]
] |
2505.12703 | Jiabin Chen | Jiabin Chen, Haiping Wang, Jinpeng Li, Yuan Liu, Zhen Dong, Bisheng Yang | SpatialLLM: From Multi-modality Data to Urban Spatial Intelligence | null | null | null | null | cs.CV cs.AI | http://creativecommons.org/licenses/by/4.0/ | We propose SpatialLLM, a novel approach advancing spatial intelligence tasks in complex urban scenes. Unlike previous methods requiring geographic analysis tools or domain expertise, SpatialLLM is a unified language model directly addressing various spatial intelligence tasks without any training, fine-tuning, or expert intervention. The core of SpatialLLM lies in constructing detailed and structured scene descriptions from raw spatial data to prompt pre-trained LLMs for scene-based analysis. Extensive experiments show that, with our designs, pretrained LLMs can accurately perceive spatial distribution information and enable zero-shot execution of advanced spatial intelligence tasks, including urban planning, ecological analysis, traffic management, etc. We argue that multi-field knowledge, context length, and reasoning ability are key factors influencing LLM performances in urban analysis. We hope that SpatialLLM will provide a novel viable perspective for urban intelligent analysis and management. The code and dataset are available at https://github.com/WHU-USI3DV/SpatialLLM. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 04:53:41 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Chen",
"Jiabin",
""
],
[
"Wang",
"Haiping",
""
],
[
"Li",
"Jinpeng",
""
],
[
"Liu",
"Yuan",
""
],
[
"Dong",
"Zhen",
""
],
[
"Yang",
"Bisheng",
""
]
] |
2505.12716 | Taiqiang Wu | Taiqiang Wu, Runming Yang, Jiayi Li, Pengfei Hu, Ngai Wong, Yujiu Yang | Shadow-FT: Tuning Instruct via Base | 19 pages, 10 tables, 6 figures | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Large language models (LLMs) consistently benefit from further fine-tuning on various tasks. However, we observe that directly tuning the INSTRUCT (i.e., instruction tuned) models often leads to marginal improvements and even performance degeneration. Notably, paired BASE models, the foundation for these INSTRUCT variants, contain highly similar weight values (i.e., less than 2% on average for Llama 3.1 8B). Therefore, we propose a novel Shadow-FT framework to tune the INSTRUCT models by leveraging the corresponding BASE models. The key insight is to fine-tune the BASE model, and then directly graft the learned weight updates to the INSTRUCT model. Our proposed Shadow-FT introduces no additional parameters, is easy to implement, and significantly improves performance. We conduct extensive experiments on tuning mainstream LLMs, such as Qwen 3 and Llama 3 series, and evaluate them across 19 benchmarks covering coding, reasoning, and mathematical tasks. Experimental results demonstrate that Shadow-FT consistently outperforms conventional full-parameter and parameter-efficient tuning approaches. Further analyses indicate that Shadow-FT can be applied to multimodal large language models (MLLMs) and combined with direct preference optimization (DPO). Codes and weights are available at \href{https://github.com/wutaiqiang/Shadow-FT}{Github}. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 05:16:21 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 03:27:18 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wu",
"Taiqiang",
""
],
[
"Yang",
"Runming",
""
],
[
"Li",
"Jiayi",
""
],
[
"Hu",
"Pengfei",
""
],
[
"Wong",
"Ngai",
""
],
[
"Yang",
"Yujiu",
""
]
] |
2505.12786 | Minrui Xu | Minrui Xu, Jiani Fan, Xinyu Huang, Conghao Zhou, Jiawen Kang, Dusit Niyato, Shiwen Mao, Zhu Han, Xuemin (Sherman) Shen, and Kwok-Yan Lam | Forewarned is Forearmed: A Survey on Large Language Model-based Agents in Autonomous Cyberattacks | null | null | null | null | cs.NI | http://creativecommons.org/licenses/by/4.0/ | With the continuous evolution of Large Language Models (LLMs), LLM-based agents have advanced beyond passive chatbots to become autonomous cyber entities capable of performing complex tasks, including web browsing, malicious code and deceptive content generation, and decision-making. By significantly reducing the time, expertise, and resources, AI-assisted cyberattacks orchestrated by LLM-based agents have led to a phenomenon termed Cyber Threat Inflation, characterized by a significant reduction in attack costs and a tremendous increase in attack scale. To provide actionable defensive insights, in this survey, we focus on the potential cyber threats posed by LLM-based agents across diverse network systems. Firstly, we present the capabilities of LLM-based cyberattack agents, which include executing autonomous attack strategies, comprising scouting, memory, reasoning, and action, and facilitating collaborative operations with other agents or human operators. Building on these capabilities, we examine common cyberattacks initiated by LLM-based agents and compare their effectiveness across different types of networks, including static, mobile, and infrastructure-free paradigms. Moreover, we analyze threat bottlenecks of LLM-based agents across different network infrastructures and review their defense methods. Due to operational imbalances, existing defense methods are inadequate against autonomous cyberattacks. Finally, we outline future research directions and potential defensive strategies for legacy network systems. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 07:19:06 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 05:00:23 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Xu",
"Minrui",
"",
"Sherman"
],
[
"Fan",
"Jiani",
"",
"Sherman"
],
[
"Huang",
"Xinyu",
"",
"Sherman"
],
[
"Zhou",
"Conghao",
"",
"Sherman"
],
[
"Kang",
"Jiawen",
"",
"Sherman"
],
[
"Niyato",
"Dusit",
"",
"Sherman"
],
[
"Mao",
"Shiwen",
"",
"Sherman"
],
[
"Han",
"Zhu",
"",
"Sherman"
],
[
"Xuemin",
"",
"",
"Sherman"
],
[
"Shen",
"",
""
],
[
"Lam",
"Kwok-Yan",
""
]
] |
2505.12795 | Shibo Hong | Shibo Hong, Jiahao Ying, Haiyuan Liang, Mengdi Zhang, Jun Kuang, Jiazheng Zhang, Yixin Cao | FRABench and GenEval: Scaling Fine-Grained Aspect Evaluation across Tasks, Modalities | null | null | null | null | cs.AI cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Evaluating the open-ended outputs of large language models (LLMs) has become a bottleneck as model capabilities, task diversity, and modality coverage rapidly expand. Existing "LLM-as-a-Judge" evaluators are typically narrow in a few tasks, aspects, or modalities, and easily suffer from low consistency. In this paper, we argue that explicit, fine-grained aspect specification is the key to both generalizability and objectivity in automated evaluation. To this end, we propose a hierarchical aspect taxonomy encompassing 112 distinct aspects that unifies evaluation across four representative settings -- Natural Language Generation, Image Understanding, Image Generation, and Interleaved Text-and-Image Generation. Building upon this taxonomy, we create FRABench, a benchmark comprising 60.4k pairwise samples with 325k evaluation labels obtained from a combination of human and LLM annotations. FRABench provides the first large-scale, multi-modal resource for training and meta-evaluating fine-grained LMM judges. Leveraging FRABench, we develop GenEval, a fine-grained evaluator generalizable across tasks and modalities. Experiments show that GenEval (i) attains high agreement with GPT-4o and expert annotators, (ii) transfers robustly to unseen tasks and modalities, and (iii) reveals systematic weaknesses of current LMMs on evaluation. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 07:29:26 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 02:54:31 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Hong",
"Shibo",
""
],
[
"Ying",
"Jiahao",
""
],
[
"Liang",
"Haiyuan",
""
],
[
"Zhang",
"Mengdi",
""
],
[
"Kuang",
"Jun",
""
],
[
"Zhang",
"Jiazheng",
""
],
[
"Cao",
"Yixin",
""
]
] |
2505.13011 | Yubin Li | Yubin Li, Xingyu Liu, Guozhang Chen | Unveiling and Steering Connectome Organization with Interpretable Latent Variables | null | null | null | null | cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The brain's intricate connectome, a blueprint for its function, presents immense complexity, yet it arises from a compact genetic code, hinting at underlying low-dimensional organizational principles. This work bridges connectomics and representation learning to uncover these principles. We propose a framework that combines subgraph extraction from the Drosophila connectome, FlyWire, with a generative model to derive interpretable low-dimensional representations of neural circuitry. Crucially, an explainability module links these latent dimensions to specific structural features, offering insights into their functional relevance. We validate our approach by demonstrating effective graph reconstruction and, significantly, the ability to manipulate these latent codes to controllably generate connectome subgraphs with predefined properties. This research offers a novel tool for understanding brain architecture and a potential avenue for designing bio-inspired artificial neural networks. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 11:54:40 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 04:10:37 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Li",
"Yubin",
""
],
[
"Liu",
"Xingyu",
""
],
[
"Chen",
"Guozhang",
""
]
] |
2505.13026 | Fazhong Liu | Jack Chen, Fazhong Liu, Naruto Liu, Yuhan Luo, Erqu Qin, Harry Zheng, Tian Dong, Haojin Zhu, Yan Meng, Xiao Wang | Step-wise Adaptive Integration of Supervised Fine-tuning and Reinforcement Learning for Task-Specific LLMs | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) excel at mathematical reasoning and logical problem-solving. The current popular training paradigms primarily use supervised fine-tuning (SFT) and reinforcement learning (RL) to enhance the models' reasoning abilities. However, when using SFT or RL alone, there are respective challenges: SFT may suffer from overfitting, while RL is prone to mode collapse. The state-of-the-art methods have proposed hybrid training schemes. However, static switching faces challenges such as poor generalization across different tasks and high dependence on data quality. In response to these challenges, inspired by the curriculum learning-quiz mechanism in human reasoning cultivation, We propose SASR, a step-wise adaptive hybrid training framework that theoretically unifies SFT and RL and dynamically balances the two throughout optimization. SASR uses SFT for initial warm-up to establish basic reasoning skills, and then uses an adaptive dynamic adjustment algorithm based on gradient norm and divergence relative to the original distribution to seamlessly integrate SFT with the online RL method GRPO. By monitoring the training status of LLMs and adjusting the training process in sequence, SASR ensures a smooth transition between training schemes, maintaining core reasoning abilities while exploring different paths. Experimental results demonstrate that SASR outperforms SFT, RL, and static hybrid training methods. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 12:10:17 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:16:10 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Chen",
"Jack",
""
],
[
"Liu",
"Fazhong",
""
],
[
"Liu",
"Naruto",
""
],
[
"Luo",
"Yuhan",
""
],
[
"Qin",
"Erqu",
""
],
[
"Zheng",
"Harry",
""
],
[
"Dong",
"Tian",
""
],
[
"Zhu",
"Haojin",
""
],
[
"Meng",
"Yan",
""
],
[
"Wang",
"Xiao",
""
]
] |
2505.13089 | Sondre Wold | Sondre Wold, Lucas Georges Gabriel Charpentier, \'Etienne Simon | Systematic Generalization in Language Models Scales with Information Entropy | Accepted to ACL 2025: Findings | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Systematic generalization remains challenging for current language models, which are known to be both sensitive to semantically similar permutations of the input and to struggle with known concepts presented in novel contexts. Although benchmarks exist for assessing compositional behavior, it is unclear how to measure the difficulty of a systematic generalization problem. In this work, we show how one aspect of systematic generalization can be described by the entropy of the distribution of component parts in the training data. We formalize a framework for measuring entropy in a sequence-to-sequence task and find that the performance of popular model architectures scales with the entropy. Our work connects systematic generalization to information efficiency, and our results indicate that success at high entropy can be achieved even without built-in priors, and that success at low entropy can serve as a target for assessing progress towards robust systematic generalization. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 13:23:44 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 05:40:05 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wold",
"Sondre",
""
],
[
"Charpentier",
"Lucas Georges Gabriel",
""
],
[
"Simon",
"Étienne",
""
]
] |
2505.13539 | Pablo Su\'arez-Serrato | Rodrigo Fritz, Pablo Su\'arez-Serrato, Victor Mijangos, Anayanzi D. Martinez-Hernandez, Eduardo Ivan Velazquez Richards | EuLearn: A 3D database for learning Euler characteristics | 35 pages, many figures. Datasets and source code publicly available at https://huggingface.co/datasets/appliedgeometry/EuLearn and https://github.com/appliedgeometry/EuLearn_db | null | null | null | cs.CG cs.CV cs.LG math.DG math.GT | http://creativecommons.org/licenses/by-nc-nd/4.0/ | We present EuLearn, the first surface datasets equitably representing a diversity of topological types. We designed our embedded surfaces of uniformly varying genera relying on random knots, thus allowing our surfaces to knot with themselves. EuLearn contributes new topological datasets of meshes, point clouds, and scalar fields in 3D. We aim to facilitate the training of machine learning systems that can discern topological features. We experimented with specific emblematic 3D neural network architectures, finding that their vanilla implementations perform poorly on genus classification. To enhance performance, we developed a novel, non-Euclidean, statistical sampling method adapted to graph and manifold data. We also introduce adjacency-informed adaptations of PointNet and Transformer architectures that rely on our non-Euclidean sampling strategy. Our results demonstrate that incorporating topological information into deep learning workflows significantly improves performance on these otherwise challenging EuLearn datasets. | [
{
"version": "v1",
"created": "Sun, 18 May 2025 19:22:04 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Fritz",
"Rodrigo",
""
],
[
"Suárez-Serrato",
"Pablo",
""
],
[
"Mijangos",
"Victor",
""
],
[
"Martinez-Hernandez",
"Anayanzi D.",
""
],
[
"Richards",
"Eduardo Ivan Velazquez",
""
]
] |
2505.13570 | Donlapark Ponnoprat | Donlapark Ponnoprat, Masaaki Imaizumi | Minimax Rates of Estimation for Optimal Transport Map between Infinite-Dimensional Spaces | 60 pages, 5 figures | null | null | null | math.ST stat.ML stat.TH | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We investigate the estimation of an optimal transport map between probability measures on an infinite-dimensional space and reveal its minimax optimal rate. Optimal transport theory defines distances within a space of probability measures, utilizing an optimal transport map as its key component. Estimating the optimal transport map from samples finds several applications, such as simulating dynamics between probability measures and functional data analysis. However, some transport maps on infinite-dimensional spaces require exponential-order data for estimation, which undermines their applicability. In this paper, we investigate the estimation of an optimal transport map between infinite-dimensional spaces, focusing on optimal transport maps characterized by the notion of $\gamma$-smoothness. Consequently, we show that the order of the minimax risk is polynomial rate in the sample size even in the infinite-dimensional setup. We also develop an estimator whose estimation error matches the minimax optimal rate. With these results, we obtain a class of reasonably estimable optimal transport maps on infinite-dimensional spaces and a method for their estimation. Our experiments validate the theory and practical utility of our approach with application to functional data analysis. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 13:00:29 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 11:11:13 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Ponnoprat",
"Donlapark",
""
],
[
"Imaizumi",
"Masaaki",
""
]
] |
2505.13577 | Taehan Kim | Yubin Kim, Taehan Kim, Wonjune Kang, Eugene Park, Joonsik Yoon, Dongjae Lee, Xin Liu, Daniel McDuff, Hyeonhoon Lee, Cynthia Breazeal, Hae Won Park | VocalAgent: Large Language Models for Vocal Health Diagnostics with Safety-Aware Evaluation | null | null | null | null | cs.SD cs.AI eess.AS | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Vocal health plays a crucial role in peoples' lives, significantly impacting their communicative abilities and interactions. However, despite the global prevalence of voice disorders, many lack access to convenient diagnosis and treatment. This paper introduces VocalAgent, an audio large language model (LLM) to address these challenges through vocal health diagnosis. We leverage Qwen-Audio-Chat fine-tuned on three datasets collected in-situ from hospital patients, and present a multifaceted evaluation framework encompassing a safety assessment to mitigate diagnostic biases, cross-lingual performance analysis, and modality ablation studies. VocalAgent demonstrates superior accuracy on voice disorder classification compared to state-of-the-art baselines. Its LLM-based method offers a scalable solution for broader adoption of health diagnostics, while underscoring the importance of ethical and technical validation. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 14:58:42 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 23:31:11 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Kim",
"Yubin",
""
],
[
"Kim",
"Taehan",
""
],
[
"Kang",
"Wonjune",
""
],
[
"Park",
"Eugene",
""
],
[
"Yoon",
"Joonsik",
""
],
[
"Lee",
"Dongjae",
""
],
[
"Liu",
"Xin",
""
],
[
"McDuff",
"Daniel",
""
],
[
"Lee",
"Hyeonhoon",
""
],
[
"Breazeal",
"Cynthia",
""
],
[
"Park",
"Hae Won",
""
]
] |
2505.13697 | Durgesh Kalwar | Soumya Rani Samineni, Durgesh Kalwar, Karthik Valmeekam, Kaya Stechly, Subbarao Kambhampati | RL in Name Only? Analyzing the Structural Assumptions in RL post-training for LLMs | null | null | null | null | cs.LG cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Reinforcement learning-based post-training of large language models (LLMs) has recently gained attention, particularly following the release of DeepSeek R1, which applied GRPO for fine-tuning. Amid the growing hype around improved reasoning abilities attributed to RL post-training, we critically examine the formulation and assumptions underlying these methods. We start by highlighting the popular structural assumptions made in modeling LLM training as a Markov Decision Process (MDP), and show how they lead to a degenerate MDP that doesn't quite need the RL/GRPO apparatus. The two critical structural assumptions include (1) making the MDP states be just a concatenation of the actions-with states becoming the context window and the actions becoming the tokens in LLMs and (2) splitting the reward of a state-action trajectory uniformly across the trajectory. Through a comprehensive analysis, we demonstrate that these simplifying assumptions make the approach effectively equivalent to an outcome-driven supervised learning. Our experiments on benchmarks including GSM8K and Countdown using Qwen-2.5 base models show that iterative supervised fine-tuning, incorporating both positive and negative samples, achieves performance comparable to GRPO-based training. We will also argue that the structural assumptions indirectly incentivize the RL to generate longer sequences of intermediate tokens-which in turn feeds into the narrative of "RL generating longer thinking traces." While RL may well be a very useful technique for improving the reasoning abilities of LLMs, our analysis shows that the simplistic structural assumptions made in modeling the underlying MDP render the popular LLM RL frameworks and their interpretations questionable. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 19:57:15 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 21:46:31 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Samineni",
"Soumya Rani",
""
],
[
"Kalwar",
"Durgesh",
""
],
[
"Valmeekam",
"Karthik",
""
],
[
"Stechly",
"Kaya",
""
],
[
"Kambhampati",
"Subbarao",
""
]
] |
2505.13710 | Noam Avidan | Noam Avidan and Rotem Arnon | Quantum Computational Unpredictability Entropy and Quantum Leakage Resilience | 41 pages. Comments welcome | null | null | null | quant-ph | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Computational entropies provide a framework for quantifying uncertainty and randomness under computational constraints. They play a central role in classical cryptography, underpinning the analysis and construction of primitives such as pseudo-random generators, leakage-resilient cryptography, and randomness extractors. In the quantum setting, however, computational analogues of entropy remain largely unexplored. In this work, we initiate the study of quantum computational entropy by defining quantum computational unpredictability entropy, a natural generalization of classical unpredictability entropy to the quantum setting. Our definition builds on the operational interpretation of quantum min-entropy as the optimal guessing probability, while restricting the adversary to efficient guessing strategies. We prove that this entropy satisfies several fundamental properties, including a leakage chain rule that holds even in the presence of unbounded prior quantum side-information. We also show that unpredictability entropy enables pseudo-randomness extraction against quantum adversaries with bounded computational power. Together, these results lay a foundation for developing cryptographic tools that rely on min-entropy in the quantum computational setting. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 20:20:30 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Avidan",
"Noam",
""
],
[
"Arnon",
"Rotem",
""
]
] |
2505.13775 | Karthik Valmeekam | Kaya Stechly, Karthik Valmeekam, Atharva Gundawar, Vardhan Palod, Subbarao Kambhampati | Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens | null | null | null | null | cs.LG cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent impressive results from large reasoning models have been interpreted as a triumph of Chain of Thought (CoT), and especially of the process of training on CoTs sampled from base LLMs in order to help find new reasoning patterns. In this paper, we critically examine that interpretation by investigating how the semantics of intermediate tokens-often anthropomorphized as "thoughts" or reasoning traces and which are claimed to display behaviors like backtracking, self-verification etc.-actually influence model performance. We train transformer models on formally verifiable reasoning traces and solutions, constraining both intermediate steps and final outputs to align with those of a formal solver (in our case, A* search). By constructing a formal interpreter of the semantics of our problems and intended algorithm, we systematically evaluate not only solution accuracy but also the correctness of intermediate traces, thus allowing us to evaluate whether the latter causally influences the former. We notice that, despite significant improvements on the solution-only baseline, models trained on entirely correct traces still produce invalid reasoning traces when arriving at correct solutions. To further show that trace accuracy is only loosely connected to solution accuracy, we then train models on noisy, corrupted traces which have no relation to the specific problem each is paired with, and find that not only does performance remain largely consistent with models trained on correct data, but in some cases can improve upon it and generalize more robustly on out-of-distribution tasks. These results challenge the assumption that intermediate tokens or "Chains of Thought" induce predictable reasoning behaviors and caution against anthropomorphizing such outputs or over-interpreting them (despite their mostly correct forms) as evidence of human-like or algorithmic behaviors in language models. | [
{
"version": "v1",
"created": "Mon, 19 May 2025 23:29:23 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:54:11 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Stechly",
"Kaya",
""
],
[
"Valmeekam",
"Karthik",
""
],
[
"Gundawar",
"Atharva",
""
],
[
"Palod",
"Vardhan",
""
],
[
"Kambhampati",
"Subbarao",
""
]
] |
2505.13797 | B Farley | B. Farley, U. T. Ahmed, A. M. Hopkins, M. Cowley, A. Battisti, S. Casura, Y. Gordon, B. W. Holwerda, S. Phillipps, C. Robertson, and T. Zafar | Galaxy And Mass Assembly: A new approach to quantifying dust in galaxies | 19 pages, 17 figures, to be published in PASA | null | 10.1017/pasa.2025.10028 | null | astro-ph.GA | http://creativecommons.org/licenses/by/4.0/ | We introduce a new approach to quantifying dust in galaxies by combining information from the Balmer decrement (BD) and the dust mass ($M_d$). While there is no explicit correlation between these two properties, they jointly probe different aspects of the dust present in galaxies. We explore two new parameters that link BD with $M_d$ by using star formation rate sensitive luminosities at several wavelengths (ultraviolet, H$\alpha$, and far-infrared). This analysis shows that combining the BD and $M_d$ in these ways provides new metrics that are sensitive to the degree of optically thick dust affecting the short wavelength emission. We show how these new ''dust geometry'' parameters vary as a function of galaxy mass, star formation rate, and specific star formation rate. We demonstrate that they are sensitive probes of the dust geometry in galaxies, and that they support the ''maximal foreground screen'' model for dust in starburst galaxies. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 01:10:40 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Farley",
"B.",
""
],
[
"Ahmed",
"U. T.",
""
],
[
"Hopkins",
"A. M.",
""
],
[
"Cowley",
"M.",
""
],
[
"Battisti",
"A.",
""
],
[
"Casura",
"S.",
""
],
[
"Gordon",
"Y.",
""
],
[
"Holwerda",
"B. W.",
""
],
[
"Phillipps",
"S.",
""
],
[
"Robertson",
"C.",
""
],
[
"Zafar",
"T.",
""
]
] |
2505.13815 | Zexin Pan | Zexin Pan | Dimension-independent convergence rates of randomized nets using median-of-means | null | null | null | null | math.NA cs.NA | http://creativecommons.org/licenses/by/4.0/ | Recent advances in quasi-Monte Carlo integration demonstrate that the median of linearly scrambled digital net estimators achieves near-optimal convergence rates for high-dimensional integrals without requiring a priori knowledge of the integrand's smoothness. Building on this framework, we prove that the median estimator attains dimension-independent convergence under tractability conditions characterized by low effective dimensionality, a property known as strong tractability in complexity theory. Our analysis strengthens existing guarantees by improving the convergence rates and relaxing the theoretical assumptions previously required for dimension-independent convergence. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 01:52:35 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 21:25:52 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Pan",
"Zexin",
""
]
] |
2505.13831 | Zongyuan Deng | Zongyuan Deng, Yujie Cai, Qing Liu, Shiyao Mu, Bin Lyu, and Zhen Yang | TelePlanNet: An AI-Driven Framework for Efficient Telecom Network Planning | 6 pages, 5 figures, 1 table | null | null | null | cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The selection of base station sites is a critical challenge in 5G network planning, which requires efficient optimization of coverage, cost, user satisfaction, and practical constraints. Traditional manual methods, reliant on human expertise, suffer from inefficiencies and are limited to an unsatisfied planning-construction consistency. Existing AI tools, despite improving efficiency in certain aspects, still struggle to meet the dynamic network conditions and multi-objective needs of telecom operators' networks. To address these challenges, we propose TelePlanNet, an AI-driven framework tailored for the selection of base station sites, integrating a three-layer architecture for efficient planning and large-scale automation. By leveraging large language models (LLMs) for real-time user input processing and intent alignment with base station planning, combined with training the planning model using the improved group relative policy optimization (GRPO) reinforcement learning, the proposed TelePlanNet can effectively address multi-objective optimization, evaluates candidate sites, and delivers practical solutions. Experiments results show that the proposed TelePlanNet can improve the consistency to 78%, which is superior to the manual methods, providing telecom operators with an efficient and scalable tool that significantly advances cellular network planning. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 02:19:10 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:33:45 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Deng",
"Zongyuan",
""
],
[
"Cai",
"Yujie",
""
],
[
"Liu",
"Qing",
""
],
[
"Mu",
"Shiyao",
""
],
[
"Lyu",
"Bin",
""
],
[
"Yang",
"Zhen",
""
]
] |
2505.13880 | Ziqian Wang | Ziqian Wang, Xianjun Xia, Xinfa Zhu, Lei Xie | U-SAM: An audio language Model for Unified Speech, Audio, and Music Understanding | Accepted to Interspeech 2025 | null | null | null | eess.AS cs.SD eess.SP | http://creativecommons.org/licenses/by/4.0/ | The text generation paradigm for audio tasks has opened new possibilities for unified audio understanding. However, existing models face significant challenges in achieving a comprehensive understanding across diverse audio types, such as speech, general audio events, and music. Furthermore, their exclusive reliance on cross-entropy loss for alignment often falls short, as it treats all tokens equally and fails to account for redundant audio features, leading to weaker cross-modal alignment. To deal with the above challenges, this paper introduces U-SAM, an advanced audio language model that integrates specialized encoders for speech, audio, and music with a pre-trained large language model (LLM). U-SAM employs a Mixture of Experts (MoE) projector for task-aware feature fusion, dynamically routing and integrating the domain-specific encoder outputs. Additionally, U-SAM incorporates a Semantic-Aware Contrastive Loss Module, which explicitly identifies redundant audio features under language supervision and rectifies their semantic and spectral representations to enhance cross-modal alignment. Extensive experiments demonstrate that U-SAM consistently outperforms both specialized models and existing audio language models across multiple benchmarks. Moreover, it exhibits emergent capabilities on unseen tasks, showcasing its generalization potential. Code is available (https://github.com/Honee-W/U-SAM/). | [
{
"version": "v1",
"created": "Tue, 20 May 2025 03:34:53 GMT"
},
{
"version": "v2",
"created": "Fri, 23 May 2025 03:10:39 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 09:36:03 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wang",
"Ziqian",
""
],
[
"Xia",
"Xianjun",
""
],
[
"Zhu",
"Xinfa",
""
],
[
"Xie",
"Lei",
""
]
] |
2505.13891 | Jin Li | Jin Li, Meijin Li, Nan Yang, Li Wang, Hao Yu, Yingzhou Huang, Kai Lin, Zi-Chao Lin, Fangyu Li | The Constraints on the SGWB from Cosmic Strings by a HFGW Electromagnetic Resonance System | 21 pages, 11 figures | null | null | null | gr-qc | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | As one of the primary detection targets for contemporary gravitational wave (GW) observatories, the stochastic gravitational wave background (SGWB) holds significant potential for enhancing our understanding of the early universe's formation and evolution. Studies indicate that the SGWB spectrum from cosmic strings can span an extraordinarily broad frequency range, extending from extremely low frequencies up to the microwave band. This work specifically investigates the detectability of cosmic string SGWB signals in an electromagnetic (EM) resonance system at GHz frequency. We present a systematic analysis encompassing: (1) the response of high frequency gravitational waves (HFGW) in such EM resonance system. (2) the development and application of fundamental data processing protocols in the EM resonance system. Our results demonstrate that the EM system shows promising sensitivity to detect cosmic string SGWB signals with tension parameters $G\mu\geq 10^{-10}$ (the corresponding dimensionless amplitude $h \geq 10^{-33}$ at 1 GHz), while potentially establishing new constraints for $G\mu\leq 10^{-10}$ in the microwave band. These findings would complement existing multi-band SGWB observations and provide additional constraints on cosmic-string tension parameters in GHz frequency regimes. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 03:55:00 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 09:04:09 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Li",
"Jin",
""
],
[
"Li",
"Meijin",
""
],
[
"Yang",
"Nan",
""
],
[
"Wang",
"Li",
""
],
[
"Yu",
"Hao",
""
],
[
"Huang",
"Yingzhou",
""
],
[
"Lin",
"Kai",
""
],
[
"Lin",
"Zi-Chao",
""
],
[
"Li",
"Fangyu",
""
]
] |
2505.13971 | Ming Gao | Ming Gao, Shilong Wu, Hang Chen, Jun Du, Chin-Hui Lee, Shinji Watanabe, Jingdong Chen, Siniscalchi Sabato Marco, Odette Scharenborg | The Multimodal Information Based Speech Processing (MISP) 2025 Challenge: Audio-Visual Diarization and Recognition | Accepted by Interspeech 2025. Camera-ready version | null | null | null | cs.SD cs.AI eess.AS | http://creativecommons.org/licenses/by/4.0/ | Meetings are a valuable yet challenging scenario for speech applications due to complex acoustic conditions. This paper summarizes the outcomes of the MISP 2025 Challenge, hosted at Interspeech 2025, which focuses on multi-modal, multi-device meeting transcription by incorporating video modality alongside audio. The tasks include Audio-Visual Speaker Diarization (AVSD), Audio-Visual Speech Recognition (AVSR), and Audio-Visual Diarization and Recognition (AVDR). We present the challenge's objectives, tasks, dataset, baseline systems, and solutions proposed by participants. The best-performing systems achieved significant improvements over the baseline: the top AVSD model achieved a Diarization Error Rate (DER) of 8.09%, improving by 7.43%; the top AVSR system achieved a Character Error Rate (CER) of 9.48%, improving by 10.62%; and the best AVDR system achieved a concatenated minimum-permutation Character Error Rate (cpCER) of 11.56%, improving by 72.49%. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 06:11:51 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 05:03:46 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Gao",
"Ming",
""
],
[
"Wu",
"Shilong",
""
],
[
"Chen",
"Hang",
""
],
[
"Du",
"Jun",
""
],
[
"Lee",
"Chin-Hui",
""
],
[
"Watanabe",
"Shinji",
""
],
[
"Chen",
"Jingdong",
""
],
[
"Marco",
"Siniscalchi Sabato",
""
],
[
"Scharenborg",
"Odette",
""
]
] |
2505.13975 | Yuxuan Jiang | Yuxuan Jiang, Dawei Li and Frank Ferraro | DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Models | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | While Large Reasoning Models (LRMs) have demonstrated success in complex reasoning tasks through long chain-of-thought (CoT) reasoning, their inference often involves excessively verbose reasoning traces, resulting in substantial inefficiency. To address this, we propose Distilled Reasoning Pruning (DRP), a hybrid framework that combines inference-time pruning with tuning-based distillation, two widely used strategies for efficient reasoning. DRP uses a teacher model to perform skill-aware step decomposition and content pruning, and then distills the pruned reasoning paths into a student model, enabling it to reason both efficiently and accurately. Across several challenging mathematical reasoning datasets, we find that models trained with DRP achieve substantial improvements in token efficiency without sacrificing accuracy. Specifically, DRP reduces average token usage on GSM8K from 917 to 328 while improving accuracy from 91.7% to 94.1%, and achieves a 43% token reduction on AIME with no performance drop. Further analysis shows that aligning the reasoning structure of training CoTs with the student's reasoning capacity is critical for effective knowledge transfer and performance gains. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 06:15:15 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 03:40:19 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Jiang",
"Yuxuan",
""
],
[
"Li",
"Dawei",
""
],
[
"Ferraro",
"Frank",
""
]
] |
2505.14223 | Jian-Min Zhang | Jiefeng Ye, Zhigao Huang, Xianxin Wu and Jian-Min Zhang | Topological electron and phonon flat bands in novel kagome superconductor XPd5 (X=Ca, Sr, Ba) | 9 pages, 5 figures, 2 tables | null | null | null | cond-mat.supr-con cond-mat.mtrl-sci | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Fermionic and bosonic localized states induced by geometric frustration in the kagome lattice provide a distinctive research platform for investigating emergent exotic quantum phenomena in strongly correlated systems. Here, we report the discovery of coexisting electronic and phononic flat bands induced by geometric frustration in a novel kagome superconductor XPd5 (X=Ca, Sr, Ba). The electronic flat band is located around the Fermi level and possesses a nontrivial topological invariant with Z2=1. Additionaly, we identify multiple van Hove singularities (vHS) arise from the kagome Pd d orbitals with distinct dispersion and sublattice features, including conventional, higher-order vHS and p-type, m-type vHS. Specifically, our investigation of the vibrational modes of the phononic flat band reveals that its formation originates from destructive interference between adjacent kagome lattice sites with antiphase vibrational modes. A spring-mass model of phonons is established to probe the physical mechanism of the phononic flat bands. Furthermore, the calculations of electron-phonon coupling in the XPd5 reveal superconducting ground states with critical temperatures (Tc) of 4.25 K, 2.75 K, and 3.35 K for CaPd5, SrPd5, and BaPd5, respectively. This work provides a promising platform to explore the Fermion-boson many-body interplay and superconducting states, while simultaneously establishing a novel analytical framework to elucidate the origin of phononic flat bands in quantum materials. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 11:32:03 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:02:16 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Ye",
"Jiefeng",
""
],
[
"Huang",
"Zhigao",
""
],
[
"Wu",
"Xianxin",
""
],
[
"Zhang",
"Jian-Min",
""
]
] |
2505.14490 | Pietro Beri | Daniele Agostini, Pietro Beri, Franco Giovenzana, \'Angel David R\'ios Ortiz | Coble duality for Jacobian Kummer fourfolds | 42 pages. Typos fixed, bibliography updated | null | null | null | math.AG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We study projective models of generalized Kummer fourfolds via O'Grady's theta groups and the classical Coble cubic. More precisely, we establish a duality between two singular models of the generalized Kummer fourfold of a Jacobian abelian surface. We also give projective models for singular Jacobian Kummer varieties of arbitrary dimension. Along the way, we also construct a first non-natural involution on the Hilbert square of a Jacobian surface. In the appendix, we study singularities of secants of arbitrary varieties at identifiable points, following Choi, Lacini, Park and Sheridan. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 15:19:32 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 22:16:37 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Agostini",
"Daniele",
""
],
[
"Beri",
"Pietro",
""
],
[
"Giovenzana",
"Franco",
""
],
[
"Ortiz",
"Ángel David Ríos",
""
]
] |
2505.14527 | Nitish Shukla | Nitish Shukla, Arun Ross | diffDemorph: Extending Reference-Free Demorphing to Unseen Faces | null | IEEE International Conference on Image Processing (ICIP), 2025 | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | A face morph is created by combining two (or more) face images corresponding to two (or more) identities to produce a composite that successfully matches the constituent identities. Reference-free (RF) demorphing reverses this process using only the morph image, without the need for additional reference images. Previous RF demorphing methods were overly constrained, as they rely on assumptions about the distributions of training and testing morphs such as the morphing technique used, face style, and images used to create the morph. In this paper, we introduce a novel diffusion-based approach that effectively disentangles component images from a composite morph image with high visual fidelity. Our method is the first to generalize across morph techniques and face styles, beating the current state of the art by $\geq 59.46\%$ under a common training protocol across all datasets tested. We train our method on morphs created using synthetically generated face images and test on real morphs, thereby enhancing the practicality of the technique. Experiments on six datasets and two face matchers establish the utility and efficacy of our method. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 15:48:27 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:05:22 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Shukla",
"Nitish",
""
],
[
"Ross",
"Arun",
""
]
] |
2505.14662 | Zdenek Sekanina | Zdenek Sekanina | The Great Comets of 1843 and 1882 at Their Previous Return to Perihelion in the Twelfth Century: One Spectacular, the Other Dull | 20 pages, 14 figures, 9 tables | null | null | null | astro-ph.EP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | New insights into the history of C/1843 D1 and C/1882 R1, the two celebrated Kreutz sungrazers, are provided by assessing evidence on their appearance at the previous perihelion return, known as X/1106 C1 and the Chinese comet of 1138 (Ho's No. 403), respectively. The conditions differed vastly because of disparities in geocentric distance, solar elongation, and phase correction (forward scattering), all linked to the arrival times (early February vs early August). The conclusions include: the daytime sighting of the 1106 comet by Sigebert de Gembloux is consistent with expectation and so are the accounts of an exceptionally long tail observed later in twilight; the comet reached perihelion only hours before its daytime detection; the 1138 comet could have never been sighted in daylight or discovered much earlier than it actually was, in early September, one month after perihelion; at discovery, the tail is predicted to have reached elevations of 15-30 deg, while the head was only 10 deg above horizon, when observed from moderate northern latitudes; the notion that Kreutz sungrazers at perihelion between mid-May and mid-August could not be seen from the ground except possibly in daylight is misleading; the appearance of the 1138 comet's nucleus after its tidal fragmentation at perihelion is modeled on the assumption that it consisted of five major fragments (including C/1882 R1 and C/1965 S1); and unpredictable morphological changes with time in the 1882 sungrazer's split nucleus are discussed. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 17:50:33 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 16:58:16 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Sekanina",
"Zdenek",
""
]
] |
2505.14667 | Wonje Jeung | Wonje Jeung, Sangyeon Yoon, Minsuk Kahng, Albert No | SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment | 22 pages | null | null | null | cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large Reasoning Models (LRMs) have become powerful tools for complex problem solving, but their structured reasoning pathways can lead to unsafe outputs when exposed to harmful prompts. Existing safety alignment methods reduce harmful outputs but can degrade reasoning depth, leading to significant trade-offs in complex, multi-step tasks, and remain vulnerable to sophisticated jailbreak attacks. To address this, we introduce SAFEPATH, a lightweight alignment method that fine-tunes LRMs to emit a short, 8-token Safety Primer at the start of their reasoning, in response to harmful prompts, while leaving the rest of the reasoning process unsupervised. Empirical results across multiple benchmarks indicate that SAFEPATH effectively reduces harmful outputs while maintaining reasoning performance. Specifically, SAFEPATH reduces harmful responses by up to 90.0% and blocks 83.3% of jailbreak attempts in the DeepSeek-R1-Distill-Llama-8B model, while requiring 295.9x less compute than Direct Refusal and 314.1x less than SafeChain. We further introduce a zero-shot variant that requires no fine-tuning. In addition, we provide a comprehensive analysis of how existing methods in LLMs generalize, or fail, when applied to reasoning-centric models, revealing critical gaps and new directions for safer AI. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 17:54:54 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:11:42 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Jeung",
"Wonje",
""
],
[
"Yoon",
"Sangyeon",
""
],
[
"Kahng",
"Minsuk",
""
],
[
"No",
"Albert",
""
]
] |
2505.14681 | Mengru Wang | Mengru Wang, Xingyu Chen, Yue Wang, Zhiwei He, Jiahao Xu, Tian Liang, Qiuzhi Liu, Yunzhi Yao, Wenxuan Wang, Ruotian Ma, Haitao Mi, Ningyu Zhang, Zhaopeng Tu, Xiaolong Li, Dong Yu | Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training | Work in progress | null | null | null | cs.AI cs.CL cs.CV cs.IR cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Mixture-of-Experts (MoE) architectures within Large Reasoning Models (LRMs) have achieved impressive reasoning capabilities by selectively activating experts to facilitate structured cognitive processes. Despite notable advances, existing reasoning models often suffer from cognitive inefficiencies like overthinking and underthinking. To address these limitations, we introduce a novel inference-time steering methodology called Reinforcing Cognitive Experts (RICE), designed to improve reasoning performance without additional training or complex heuristics. Leveraging normalized Pointwise Mutual Information (nPMI), we systematically identify specialized experts, termed ''cognitive experts'' that orchestrate meta-level reasoning operations characterized by tokens like ''<think>''. Empirical evaluations with leading MoE-based LRMs (DeepSeek-R1 and Qwen3-235B) on rigorous quantitative and scientific reasoning benchmarks demonstrate noticeable and consistent improvements in reasoning accuracy, cognitive efficiency, and cross-domain generalization. Crucially, our lightweight approach substantially outperforms prevalent reasoning-steering techniques, such as prompt design and decoding constraints, while preserving the model's general instruction-following skills. These results highlight reinforcing cognitive experts as a promising, practical, and interpretable direction to enhance cognitive efficiency within advanced reasoning models. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 17:59:16 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 09:35:12 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wang",
"Mengru",
""
],
[
"Chen",
"Xingyu",
""
],
[
"Wang",
"Yue",
""
],
[
"He",
"Zhiwei",
""
],
[
"Xu",
"Jiahao",
""
],
[
"Liang",
"Tian",
""
],
[
"Liu",
"Qiuzhi",
""
],
[
"Yao",
"Yunzhi",
""
],
[
"Wang",
"Wenxuan",
""
],
[
"Ma",
"Ruotian",
""
],
[
"Mi",
"Haitao",
""
],
[
"Zhang",
"Ningyu",
""
],
[
"Tu",
"Zhaopeng",
""
],
[
"Li",
"Xiaolong",
""
],
[
"Yu",
"Dong",
""
]
] |
2505.14832 | Wonje Jeung | Wonje Jeung, Sangyeon Yoon, Albert No | SEPS: A Separability Measure for Robust Unlearning in LLMs | 32 pages | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Machine unlearning aims to selectively remove targeted knowledge from Large Language Models (LLMs), ensuring they forget specified content while retaining essential information. Existing unlearning metrics assess whether a model correctly answers retain queries and rejects forget queries, but they fail to capture real-world scenarios where forget queries rarely appear in isolation. In fact, forget and retain queries often coexist within the same prompt, making mixed-query evaluation crucial.
We introduce SEPS, an evaluation framework that explicitly measures a model's ability to both forget and retain information within a single prompt. Through extensive experiments across three benchmarks, we identify two key failure modes in existing unlearning methods: (1) untargeted unlearning indiscriminately erases both forget and retain content once a forget query appears, and (2) targeted unlearning overfits to single-query scenarios, leading to catastrophic failures when handling multiple queries. To address these issues, we propose Mixed Prompt (MP) unlearning, a strategy that integrates both forget and retain queries into a unified training objective. Our approach significantly improves unlearning effectiveness, demonstrating robustness even in complex settings with up to eight mixed forget and retain queries in a single prompt. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 18:53:32 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:11:40 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Jeung",
"Wonje",
""
],
[
"Yoon",
"Sangyeon",
""
],
[
"No",
"Albert",
""
]
] |
2505.14874 | Eunjung Yeo | Chin-Jou Li, Eunjung Yeo, Kwanghee Choi, Paula Andrea P\'erez-Toro, Masao Someki, Rohan Kumar Das, Zhengjun Yue, Juan Rafael Orozco-Arroyave, Elmar N\"oth, David R. Mortensen | Towards Inclusive ASR: Investigating Voice Conversion for Dysarthric Speech Recognition in Low-Resource Languages | 5 pages, 1 figure, Accepted to Interspeech 2025 | null | null | null | cs.CL cs.SD eess.AS | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Automatic speech recognition (ASR) for dysarthric speech remains challenging due to data scarcity, particularly in non-English languages. To address this, we fine-tune a voice conversion model on English dysarthric speech (UASpeech) to encode both speaker characteristics and prosodic distortions, then apply it to convert healthy non-English speech (FLEURS) into non-English dysarthric-like speech. The generated data is then used to fine-tune a multilingual ASR model, Massively Multilingual Speech (MMS), for improved dysarthric speech recognition. Evaluation on PC-GITA (Spanish), EasyCall (Italian), and SSNCE (Tamil) demonstrates that VC with both speaker and prosody conversion significantly outperforms the off-the-shelf MMS performance and conventional augmentation techniques such as speed and tempo perturbation. Objective and subjective analyses of the generated data further confirm that the generated speech simulates dysarthric characteristics. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 20:03:45 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 02:04:37 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Li",
"Chin-Jou",
""
],
[
"Yeo",
"Eunjung",
""
],
[
"Choi",
"Kwanghee",
""
],
[
"Pérez-Toro",
"Paula Andrea",
""
],
[
"Someki",
"Masao",
""
],
[
"Das",
"Rohan Kumar",
""
],
[
"Yue",
"Zhengjun",
""
],
[
"Orozco-Arroyave",
"Juan Rafael",
""
],
[
"Nöth",
"Elmar",
""
],
[
"Mortensen",
"David R.",
""
]
] |
2505.14910 | Yu Zhang | Yu Zhang, Wenxiang Guo, Changhao Pan, Dongyu Yao, Zhiyuan Zhu, Ziyue Jiang, Yuhan Wang, Tao Jin, Zhou Zhao | TCSinger 2: Customizable Multilingual Zero-shot Singing Voice Synthesis | Accepted by Findings of ACL 2025 | null | null | null | eess.AS cs.CL cs.SD | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Customizable multilingual zero-shot singing voice synthesis (SVS) has various potential applications in music composition and short video dubbing. However, existing SVS models overly depend on phoneme and note boundary annotations, limiting their robustness in zero-shot scenarios and producing poor transitions between phonemes and notes. Moreover, they also lack effective multi-level style control via diverse prompts. To overcome these challenges, we introduce TCSinger 2, a multi-task multilingual zero-shot SVS model with style transfer and style control based on various prompts. TCSinger 2 mainly includes three key modules: 1) Blurred Boundary Content (BBC) Encoder, predicts duration, extends content embedding, and applies masking to the boundaries to enable smooth transitions. 2) Custom Audio Encoder, uses contrastive learning to extract aligned representations from singing, speech, and textual prompts. 3) Flow-based Custom Transformer, leverages Cus-MOE, with F0 supervision, enhancing both the synthesis quality and style modeling of the generated singing voice. Experimental results show that TCSinger 2 outperforms baseline models in both subjective and objective metrics across multiple related tasks. Singing voice samples are available at https://aaronz345.github.io/TCSinger2Demo/. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 21:04:10 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 00:54:20 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Zhang",
"Yu",
""
],
[
"Guo",
"Wenxiang",
""
],
[
"Pan",
"Changhao",
""
],
[
"Yao",
"Dongyu",
""
],
[
"Zhu",
"Zhiyuan",
""
],
[
"Jiang",
"Ziyue",
""
],
[
"Wang",
"Yuhan",
""
],
[
"Jin",
"Tao",
""
],
[
"Zhao",
"Zhou",
""
]
] |
2505.15209 | Wonje Jeung | Wonje Jeung, Sangyeon Yoon, Hyesoo Hong, Soeun Kim, Seungju Han, Youngjae Yu, Albert No | DUSK: Do Not Unlearn Shared Knowledge | 21 pages | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) are increasingly deployed in real-world applications, raising concerns about the unauthorized use of copyrighted or sensitive data. Machine unlearning aims to remove such 'forget' data while preserving utility and information from the 'retain' set. However, existing evaluations typically assume that forget and retain sets are fully disjoint, overlooking realistic scenarios where they share overlapping content. For instance, a news article may need to be unlearned, even though the same event, such as an earthquake in Japan, is also described factually on Wikipedia. Effective unlearning should remove the specific phrasing of the news article while preserving publicly supported facts. In this paper, we introduce DUSK, a benchmark designed to evaluate unlearning methods under realistic data overlap. DUSK constructs document sets that describe the same factual content in different styles, with some shared information appearing across all sets and other content remaining unique to each. When one set is designated for unlearning, an ideal method should remove its unique content while preserving shared facts. We define seven evaluation metrics to assess whether unlearning methods can achieve this selective removal. Our evaluation of nine recent unlearning methods reveals a key limitation: while most can remove surface-level text, they often fail to erase deeper, context-specific knowledge without damaging shared content. We release DUSK as a public benchmark to support the development of more precise and reliable unlearning techniques for real-world applications. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 07:37:35 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:11:39 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Jeung",
"Wonje",
""
],
[
"Yoon",
"Sangyeon",
""
],
[
"Hong",
"Hyesoo",
""
],
[
"Kim",
"Soeun",
""
],
[
"Han",
"Seungju",
""
],
[
"Yu",
"Youngjae",
""
],
[
"No",
"Albert",
""
]
] |
2505.15214 | Sangyeon Yoon | Sangyeon Yoon, Wonje Jeung, Albert No | R-TOFU: Unlearning in Large Reasoning Models | 19 pages | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large Reasoning Models (LRMs) embed private or copyrighted information not only in their final answers but also throughout multi-step chain-of-thought (CoT) traces, making reliable unlearning far more demanding than in standard LLMs. We introduce Reasoning-TOFU (R-TOFU), the first benchmark tailored to this setting. R-TOFU augments existing unlearning tasks with realistic CoT annotations and provides step-wise metrics that expose residual knowledge invisible to answer-level checks. Using R-TOFU, we carry out a comprehensive comparison of gradient-based and preference-optimization baselines and show that conventional answer-only objectives leave substantial forget traces in reasoning. We further propose Reasoned IDK, a preference-optimization variant that preserves coherent yet inconclusive reasoning, achieving a stronger balance between forgetting efficacy and model utility than earlier refusal styles. Finally, we identify a failure mode: decoding variants such as ZeroThink and LessThink can still reveal forgotten content despite seemingly successful unlearning, emphasizing the need to evaluate models under diverse decoding settings. Together, the benchmark, analysis, and new baseline establish a systematic foundation for studying and improving unlearning in LRMs while preserving their reasoning capabilities. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 07:44:30 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:37:47 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Yoon",
"Sangyeon",
""
],
[
"Jeung",
"Wonje",
""
],
[
"No",
"Albert",
""
]
] |
2505.15357 | Subhash Chandra Mahapatra | Nikesh Lilani, Dilpreet Sandhu, Subhash Mahapatra | Comparative study of the butterfly velocity in holographic QCD models at finite temperature and chemical potential | 40 pages, 9 figures, author name and typos corrected, discussion on technical issues of the numerical model added, references added | null | null | null | hep-th hep-lat hep-ph nlin.CD quant-ph | http://creativecommons.org/licenses/by/4.0/ | In this work, we study quantum chaos in a variety of holographic QCD models at finite temperature and chemical potentials. This includes the 1 R-Charge black hole (1RCBH) model, the 2 R-Charge black hole (2RCBH) model, a potential reconstruction-based analytic bottom-up model, and a numerical bottom-up model. All these models are different avatars of the Einstein-Maxwell-dilaton gravity action, distinguished by their specific choices of dilaton potentials and gauge-kinetic coupling functions. We focus on computing the chaos parameter, the butterfly velocity, using three distinct methods: entanglement wedge reconstruction, out-of-time-ordered correlators (OTOCs), and pole-skipping. We show that all three methods yield identical results for the butterfly velocity across all the holographic QCD models considered, further establishing the equivalence between the three approaches. Furthermore, we analyze in detail the behavior of the butterfly velocity as a function of chemical potential and temperature. Interestingly, a universal trend emerges across all models: the butterfly velocity increases/decreases with temperature/chemical potential for thermodynamically stable phases. Additionally, in the high-temperature limit, the butterfly velocity in all models approaches that of the chargeless plasma. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 10:40:11 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 05:00:46 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Lilani",
"Nikesh",
""
],
[
"Sandhu",
"Dilpreet",
""
],
[
"Mahapatra",
"Subhash",
""
]
] |
2505.15536 | Jinquan Wang | Jinquan Wang, Xiaojian Liao, Xuzhao Liu, Jiashun Suo, Zhisheng Huo, Chenhao Zhang, Xiangrong Xu, Runnan Shen, Xilong Xie, Limin Xiao | DeepCEE: Efficient Cross-Region Model Distributed Training System under Heterogeneous GPUs and Networks | null | null | null | null | eess.SY cs.DC cs.SY | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Most existing training systems focus on a single region. In contrast, we envision that cross-region training offers more flexible GPU resource allocation and yields significant potential. However, the hierarchical cluster topology and unstable networks in the cloud-edge-end (CEE) environment, a typical cross-region scenario, pose substantial challenges to building an efficient and autonomous model training system. We propose DeepCEE, a geo-distributed model training system tailored for heterogeneous GPUs and networks in CEE environments. DeepCEE adopts a communication-centric design philosophy to tackle challenges arising from slow and unstable inter-region networks. It begins with a heterogeneous device profiler that identifies and groups devices based on both network and compute characteristics. Leveraging device groups, DeepCEE implements compact, zero-bubble pipeline parallelism, automatically deriving optimal parallel strategies. To further adapt to runtime variability, DeepCEE integrates a dynamic environment adapter that reacts to network fluctuations. Extensive evaluations demonstrate that DeepCEE achieves 1.3-2.8x higher training throughput compared to widely used and SOTA training systems. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 13:58:13 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:35:57 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wang",
"Jinquan",
""
],
[
"Liao",
"Xiaojian",
""
],
[
"Liu",
"Xuzhao",
""
],
[
"Suo",
"Jiashun",
""
],
[
"Huo",
"Zhisheng",
""
],
[
"Zhang",
"Chenhao",
""
],
[
"Xu",
"Xiangrong",
""
],
[
"Shen",
"Runnan",
""
],
[
"Xie",
"Xilong",
""
],
[
"Xiao",
"Limin",
""
]
] |
2505.15599 | Rohit Sarma Sarkar | Aniket Basak, Rajeet Ghosh, Rohit Sarma Sarkar, Chandan Goswami, Avishek Adhikari | Device-Independent Ternary Quantum Key Distribution Protocol Based on the Impossible Colouring Game | 25 pages, 3 figures | null | null | null | quant-ph | http://creativecommons.org/licenses/by-nc-sa/4.0/ | We propose a Ternary Device-Independent Quantum Key Distribution (TDIQKD) protocol based on the two-party Impossible Colouring pseudo-telepathy game, using maximally entangled qutrit states to enable secure key generation between distant parties. The protocol makes use of non-contextuality from the Kochen-Specker theorem, providing a quantum advantage in a task that is classically unachievable, and eliminates reliance on assumptions about the internal workings of quantum devices. A specially designed qutrit quantum circuit is employed for state preparation, and security is rigorously analyzed through a composable framework using smooth minimum entropy, von Neumann entropy, and Shannon entropy. The protocol achieves optimal key rate in the ideal case and maintains security under significant noise, with a finite-key analysis supporting practical deployment. The protocol is secure within an adequate security framework and demonstrates a higher key generation rate compared to standard QKD protocols, emphasizing the potential of high-dimensional quantum systems for secure communication. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 14:52:31 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 18:26:58 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Basak",
"Aniket",
""
],
[
"Ghosh",
"Rajeet",
""
],
[
"Sarkar",
"Rohit Sarma",
""
],
[
"Goswami",
"Chandan",
""
],
[
"Adhikari",
"Avishek",
""
]
] |
2505.15617 | Arsene Brice Zotsa Ngoufack | Arsene Brice Zotsa-Ngoufack | Functional Central Limit Theorem and SPDE for epidemic model with memory of the last infection and waning immunity | null | null | null | null | math.PR | http://creativecommons.org/licenses/by/4.0/ | We study the fluctuations of a stochastic epidemic model with memory of the last infections, varying infectivity, and waning immunity, as introduced in Guerin and Zotsa-Ngoufack:arXiv preprint arXiv:2505.00601. The dynamics of the epidemic model are described by a measure-valued process with respect to infection age and individual traits. The Functional Law of Large Numbers (FLLN) is formulated as an integral equation, which is solved by a deterministic measure. In this article, we establish the Functional Central Limit Theorem (FCLT), capturing the fluctuations of the stochastic model around its deterministic limit. The limit of the FCLT is given by a nonlinear stochastic integral equation which is solved by a random signed-measure. We further derive the weak solution in the form of a stochastic partial differential equation (SPDE) and propose an alternative representation of the FCLT, as fluctuations in the average total force of infection and average susceptibility. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 15:07:07 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 11:29:47 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Zotsa-Ngoufack",
"Arsene Brice",
""
]
] |
2505.15812 | Satoshi Kosugi | Satoshi Kosugi | Leveraging the Powerful Attention of a Pre-trained Diffusion Model for Exemplar-based Image Colorization | Accepted to IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) | null | 10.1109/TCSVT.2025.3573401 | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Exemplar-based image colorization aims to colorize a grayscale image using a reference color image, ensuring that reference colors are applied to corresponding input regions based on their semantic similarity. To achieve accurate semantic matching between regions, we leverage the self-attention module of a pre-trained diffusion model, which is trained on a large dataset and exhibits powerful attention capabilities. To harness this power, we propose a novel, fine-tuning-free approach based on a pre-trained diffusion model, making two key contributions. First, we introduce dual attention-guided color transfer. We utilize the self-attention module to compute an attention map between the input and reference images, effectively capturing semantic correspondences. The color features from the reference image is then transferred to the semantically matching regions of the input image, guided by this attention map, and finally, the grayscale features are replaced with the corresponding color features. Notably, we utilize dual attention to calculate attention maps separately for the grayscale and color images, achieving more precise semantic alignment. Second, we propose classifier-free colorization guidance, which enhances the transferred colors by combining color-transferred and non-color-transferred outputs. This process improves the quality of colorization. Our experimental results demonstrate that our method outperforms existing techniques in terms of image quality and fidelity to the reference. Specifically, we use 335 input-reference pairs from previous research, achieving an FID of 95.27 (image quality) and an SI-FID of 5.51 (fidelity to the reference). Our source code is available at https://github.com/satoshi-kosugi/powerful-attention. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 17:59:40 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Kosugi",
"Satoshi",
""
]
] |
2505.15974 | Alan Ta | Alan Ta, Nilsu Salgin, Mustafa Demir, Kala Phillips Reindel, Ranjana K. Mehta, Anthony McDonald, Carly McCord, Farzan Sasangohar | Real-Time Stress Monitoring, Detection, and Management in College Students: A Wearable Technology and Machine-Learning Approach | 30 pages, 5 figures | null | null | null | cs.HC cs.LG | http://creativecommons.org/licenses/by/4.0/ | College students are increasingly affected by stress, anxiety, and depression, yet face barriers to traditional mental health care. This study evaluated the efficacy of a mobile health (mHealth) intervention, Mental Health Evaluation and Lookout Program (mHELP), which integrates a smartwatch sensor and machine learning (ML) algorithms for real-time stress detection and self-management. In a 12-week randomized controlled trial (n = 117), participants were assigned to a treatment group using mHELP's full suite of interventions or a control group using the app solely for real-time stress logging and weekly psychological assessments. The primary outcome, "Moments of Stress" (MS), was assessed via physiological and self-reported indicators and analyzed using Generalized Linear Mixed Models (GLMM) approaches. Similarly, secondary outcomes of psychological assessments, including the Generalized Anxiety Disorder-7 (GAD-7) for anxiety, the Patient Health Questionnaire (PHQ-8) for depression, and the Perceived Stress Scale (PSS), were also analyzed via GLMM. The finding of the objective measure, MS, indicates a substantial decrease in MS among the treatment group compared to the control group, while no notable between-group differences were observed in subjective scores of anxiety (GAD-7), depression (PHQ-8), or stress (PSS). However, the treatment group exhibited a clinically meaningful decline in GAD-7 and PSS scores. These findings underscore the potential of wearable-enabled mHealth tools to reduce acute stress in college populations and highlight the need for extended interventions and tailored features to address chronic symptoms like depression. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 19:45:42 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 01:50:01 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Ta",
"Alan",
""
],
[
"Salgin",
"Nilsu",
""
],
[
"Demir",
"Mustafa",
""
],
[
"Reindel",
"Kala Phillips",
""
],
[
"Mehta",
"Ranjana K.",
""
],
[
"McDonald",
"Anthony",
""
],
[
"McCord",
"Carly",
""
],
[
"Sasangohar",
"Farzan",
""
]
] |
2505.15987 | Aaron Zweig | Aaron Zweig, Zaikang Lin, Elham Azizi, David Knowles | Towards Identifiability of Interventional Stochastic Differential Equations | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | We study identifiability of stochastic differential equation (SDE) models under multiple interventions. Our results give the first provable bounds for unique recovery of SDE parameters given samples from their stationary distributions. We give tight bounds on the number of necessary interventions for linear SDEs, and upper bounds for nonlinear SDEs in the small noise regime. We experimentally validate the recovery of true parameters in synthetic data, and motivated by our theoretical results, demonstrate the advantage of parameterizations with learnable activation functions. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 20:10:54 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 13:55:15 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Zweig",
"Aaron",
""
],
[
"Lin",
"Zaikang",
""
],
[
"Azizi",
"Elham",
""
],
[
"Knowles",
"David",
""
]
] |
2505.16040 | Kazuma Ohara | Kazuma Ohara | On parameters of Hecke algebras for $p$-adic groups | 29 pages, Improved the definition of the group G_theta in Section 4.3. Added some related references | null | null | MPIM-Bonn-2025 | math.RT math.NT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Let $F$ be a non-archimedean local field with residue characteristic $p$ and $G$ be a connected reductive group defined over $F$. In earlier joint works with Jeffrey D. Adler, Jessica Fintzen, and Manish Mishra, we proved that the Hecke algebras attached to types constructed by Kim and Yu are isomorphic to the Hecke algebras attached to depth-zero types. Note that if $G$ splits over a tamely ramified extension of $F$ and $p$ does not divide the order of the absolute Weyl group of $G$, such Hecke algebras cover the Hecke algebras attached to arbitrary Bernstein blocks. We also proved that for a depth-zero type $(K, \rho)$, the corresponding Hecke algebra $\mathcal{H}(G(F), (K, \rho))$ has an explicit description as a semi-direct product of an affine Hecke algebra $\mathcal{H}(W(\rho_M)_{\mathrm{aff}}, q)$ with a twisted group algebra $\mathbb{C}[\Omega(\rho_{M}), \mu]$, generalizing prior work of Morris.
In this paper, we show that the affine Hecke algebra $\mathcal{H}(W(\rho_M)_{\mathrm{aff}}, q)$ appearing in the description of the Hecke algebra $\mathcal{H}(G(F), (K, \rho))$ attached to a depth-zero type $(K, \rho)$ is isomorphic to the one attached to a unipotent type for a connected reductive group splitting over an unramified extension of $F$. This makes it possible to calculate the parameters of the affine Hecke algebras for depth-zero types and types constructed by Kim and Yu explicitly. In particular, we prove a version of Lusztig's conjecture that the parameters of the Hecke algebra attached to an arbitrary Bernstein block agree with those of a unipotent Bernstein block under the assumption that $G$ splits over a tamely ramified extension of $F$ and $p$ does not divide the order of the absolute Weyl group of $G$. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 21:44:04 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 11:40:25 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Ohara",
"Kazuma",
""
]
] |
2505.16048 | Philipp Siedler | Philipp D. Siedler | SPhyR: Spatial-Physical Reasoning Benchmark on Material Distribution | null | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | We introduce a novel dataset designed to benchmark the physical and spatial reasoning capabilities of Large Language Models (LLM) based on topology optimization, a method for computing optimal material distributions within a design space under prescribed loads and supports. In this dataset, LLMs are provided with conditions such as 2D boundary, applied forces and supports, and must reason about the resulting optimal material distribution. The dataset includes a variety of tasks, ranging from filling in masked regions within partial structures to predicting complete material distributions. Solving these tasks requires understanding the flow of forces and the required material distribution under given constraints, without access to simulation tools or explicit physical models, challenging models to reason about structural stability and spatial organization. Our dataset targets the evaluation of spatial and physical reasoning abilities in 2D settings, offering a complementary perspective to traditional language and logic benchmarks. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 22:00:20 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 12:03:40 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Siedler",
"Philipp D.",
""
]
] |
2505.16074 | Olaoluwa Adigun | Bart Kosko, Olaoluwa Adigun | Bidirectional Variational Autoencoders | 10 pages, 6 figures | null | null | null | cs.LG cs.AI stat.ML | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present the new bidirectional variational autoencoder (BVAE) network architecture. The BVAE uses a single neural network both to encode and decode instead of an encoder-decoder network pair. The network encodes in the forward direction and decodes in the backward direction through the same synaptic web. Simulations compared BVAEs and ordinary VAEs on the four image tasks of image reconstruction, classification, interpolation, and generation. The image datasets included MNIST handwritten digits, Fashion-MNIST, CIFAR-10, and CelebA-64 face images. The bidirectional structure of BVAEs cut the parameter count by almost 50% and still slightly outperformed the unidirectional VAEs. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 23:19:43 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 23:36:25 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Kosko",
"Bart",
""
],
[
"Adigun",
"Olaoluwa",
""
]
] |
2505.16156 | Siu Lun Chau | Siu Lun Chau, Michele Caprio, Krikamol Muandet | Integral Imprecise Probability Metrics | 37 pages | null | null | null | stat.ML cs.LG | http://creativecommons.org/licenses/by/4.0/ | Quantifying differences between probability distributions is fundamental to statistics and machine learning, primarily for comparing statistical uncertainty. In contrast, epistemic uncertainty (EU) -- due to incomplete knowledge -- requires richer representations than those offered by classical probability. Imprecise probability (IP) theory offers such models, capturing ambiguity and partial belief. This has driven growing interest in imprecise probabilistic machine learning (IPML), where inference and decision-making rely on broader uncertainty models -- highlighting the need for metrics beyond classical probability. This work introduces the Integral Imprecise Probability Metric (IIPM) framework, a Choquet integral-based generalisation of classical Integral Probability Metric (IPM) to the setting of capacities -- a broad class of IP models encompassing many existing ones, including lower probabilities, probability intervals, belief functions, and more. Theoretically, we establish conditions under which IIPM serves as a valid metric and metrises a form of weak convergence of capacities. Practically, IIPM not only enables comparison across different IP models but also supports the quantification of epistemic uncertainty within a single IP model. In particular, by comparing an IP model with its conjugate, IIPM gives rise to a new class of EU measures -- Maximum Mean Imprecision -- which satisfy key axiomatic properties proposed in the Uncertainty Quantification literature. We validate MMI through selective classification experiments, demonstrating strong empirical performance against established EU measures, and outperforming them when classical methods struggle to scale to a large number of classes. Our work advances both theory and practice in IPML, offering a principled framework for comparing and quantifying epistemic uncertainty under imprecision. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 02:56:57 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 01:21:30 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Chau",
"Siu Lun",
""
],
[
"Caprio",
"Michele",
""
],
[
"Muandet",
"Krikamol",
""
]
] |
2505.16369 | Junbo Zhang | Junbo Zhang, Heinrich Dinkel, Yadong Niu, Chenyu Liu, Si Cheng, Anbei Zhao, Jian Luan | X-ARES: A Comprehensive Framework for Assessing Audio Encoder Performance | Accepted by Interspeech 2025 | null | null | null | cs.SD eess.AS | http://creativecommons.org/licenses/by/4.0/ | We introduces X-ARES (eXtensive Audio Representation and Evaluation Suite), a novel open-source benchmark designed to systematically assess audio encoder performance across diverse domains. By encompassing tasks spanning speech, environmental sounds, and music, X-ARES provides two evaluation approaches for evaluating audio representations: linear fine-tuning and unparameterized evaluation. The framework includes 22 distinct tasks that cover essential aspects of audio processing, from speech recognition and emotion detection to sound event classification and music genre identification. Our extensive evaluation of state-of-the-art audio encoders reveals significant performance variations across different tasks and domains, highlighting the complexity of general audio representation learning. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 08:23:54 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 06:49:39 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Zhang",
"Junbo",
""
],
[
"Dinkel",
"Heinrich",
""
],
[
"Niu",
"Yadong",
""
],
[
"Liu",
"Chenyu",
""
],
[
"Cheng",
"Si",
""
],
[
"Zhao",
"Anbei",
""
],
[
"Luan",
"Jian",
""
]
] |
2505.16401 | Xiaoqing Zhang | Xiaoqing Zhang, Huabin Zheng, Ang Lv, Yuhan Liu, Zirui Song, Flood Sung, Xiuying Chen, Rui Yan | Divide-Fuse-Conquer: Eliciting "Aha Moments" in Multi-Scenario Games | 25 pages, 13 figures, and 8 tables | null | null | null | cs.LG | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Large language models (LLMs) have been observed to suddenly exhibit advanced reasoning abilities during reinforcement learning (RL), resembling an ``aha moment'' triggered by simple outcome-based rewards. While RL has proven effective in eliciting such breakthroughs in tasks involving mathematics, coding, and vision, it faces significant challenges in multi-scenario games. The diversity of game rules, interaction modes, and environmental complexities often leads to policies that perform well in one scenario but fail to generalize to others. Simply combining multiple scenarios during training introduces additional challenges, such as training instability and poor performance. To overcome these challenges, we propose Divide-Fuse-Conquer, a framework designed to enhance generalization in multi-scenario RL. This approach starts by heuristically grouping games based on characteristics such as rules and difficulties. Specialized models are then trained for each group to excel at games in the group is what we refer to as the divide step. Next, we fuse model parameters from different groups as a new model, and continue training it for multiple groups, until the scenarios in all groups are conquered. Experiments across 18 TextArena games show that Qwen2.5-32B-Align trained with the Divide-Fuse-Conquer strategy reaches a performance level comparable to Claude3.5, achieving 7 wins and 4 draws. We hope our approach can inspire future research on using reinforcement learning to improve the generalization of LLMs. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 08:52:21 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 06:16:55 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Zhang",
"Xiaoqing",
""
],
[
"Zheng",
"Huabin",
""
],
[
"Lv",
"Ang",
""
],
[
"Liu",
"Yuhan",
""
],
[
"Song",
"Zirui",
""
],
[
"Sung",
"Flood",
""
],
[
"Chen",
"Xiuying",
""
],
[
"Yan",
"Rui",
""
]
] |
2505.16448 | Renfei Dang | Renfei Dang, Shujian Huang, Jiajun Chen | Internal Bias in Reasoning Models leads to Overthinking | null | null | null | null | cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | While current reasoning models possess strong exploratory capabilities, they are often criticized for overthinking due to redundant and unnecessary reflections. In this work, we reveal for the first time that overthinking in reasoning models may stem from their internal bias towards input texts. Upon encountering a reasoning problem, the model immediately forms a preliminary guess about the answer, which we term as an internal bias since it is not derived through actual reasoning. When this guess conflicts with its reasoning result, the model tends to engage in reflection, leading to the waste of computational resources. Through further interpretability experiments, we find that this behavior is largely driven by the model's excessive attention to the input section, which amplifies the influence of internal bias on its decision-making process. Additionally, by masking out the original input section, the affect of internal bias can be effectively alleviated and the reasoning length could be reduced by 31%-53% across different complex reasoning tasks. Notably, in most cases, this approach also leads to improvements in accuracy. These findings demonstrate a causal relationship between internal bias and overthinking. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 09:35:52 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 09:44:20 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Dang",
"Renfei",
""
],
[
"Huang",
"Shujian",
""
],
[
"Chen",
"Jiajun",
""
]
] |
2505.16459 | Guiyao Tie | Guiyao Tie, Xueyang Zhou, Tianhe Gu, Ruihang Zhang, Chaoran Hu, Sizhe Zhang, Mengqu Sun, Yan Zhang, Pan Zhou, Lichao Sun | MMMR: Benchmarking Massive Multi-Modal Reasoning Tasks | 39 pages, 28 figures, 4 tables | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Recent advances in Multi-Modal Large Language Models (MLLMs) have enabled unified processing of language, vision, and structured inputs, opening the door to complex tasks such as logical deduction, spatial reasoning, and scientific analysis. Despite their promise, the reasoning capabilities of MLLMs, particularly those augmented with intermediate thinking traces (MLLMs-T), remain poorly understood and lack standardized evaluation benchmarks. Existing work focuses primarily on perception or final answer correctness, offering limited insight into how models reason or fail across modalities. To address this gap, we introduce the MMMR, a new benchmark designed to rigorously evaluate multi-modal reasoning with explicit thinking. The MMMR comprises 1) a high-difficulty dataset of 1,083 questions spanning six diverse reasoning types with symbolic depth and multi-hop demands and 2) a modular Reasoning Trace Evaluation Pipeline (RTEP) for assessing reasoning quality beyond accuracy through metrics like relevance, consistency, and structured error annotations. Empirical results show that MLLMs-T overall outperform non-thinking counterparts, but even top models like Claude-3.7-Sonnet and Gemini-2.5 Pro suffer from reasoning pathologies such as inconsistency and overthinking. This benchmark reveals persistent gaps between accuracy and reasoning quality and provides an actionable evaluation pipeline for future model development. Overall, the MMMR offers a scalable foundation for evaluating, comparing, and improving the next generation of multi-modal reasoning systems. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 09:41:55 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 06:30:59 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Tie",
"Guiyao",
""
],
[
"Zhou",
"Xueyang",
""
],
[
"Gu",
"Tianhe",
""
],
[
"Zhang",
"Ruihang",
""
],
[
"Hu",
"Chaoran",
""
],
[
"Zhang",
"Sizhe",
""
],
[
"Sun",
"Mengqu",
""
],
[
"Zhang",
"Yan",
""
],
[
"Zhou",
"Pan",
""
],
[
"Sun",
"Lichao",
""
]
] |
2505.16522 | Zhouhao Sun | Zhouhao Sun, Zhiyuan Kan, Xiao Ding, Li Du, Yang Zhao, Bing Qin, Ting Liu | Benchmarking and Pushing the Multi-Bias Elimination Boundary of LLMs via Causal Effect Estimation-guided Debiasing | null | null | null | null | cs.CL cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Despite significant progress, recent studies have indicated that current large language models (LLMs) may still utilize bias during inference, leading to the poor generalizability of LLMs. Some benchmarks are proposed to investigate the generalizability of LLMs, with each piece of data typically containing one type of controlled bias. However, a single piece of data may contain multiple types of biases in practical applications. To bridge this gap, we propose a multi-bias benchmark where each piece of data contains five types of biases. The evaluations conducted on this benchmark reveal that the performance of existing LLMs and debiasing methods is unsatisfying, highlighting the challenge of eliminating multiple types of biases simultaneously. To overcome this challenge, we propose a causal effect estimation-guided multi-bias elimination method (CMBE). This method first estimates the causal effect of multiple types of biases simultaneously. Subsequently, we eliminate the causal effect of biases from the total causal effect exerted by both the semantic information and biases during inference. Experimental results show that CMBE can effectively eliminate multiple types of bias simultaneously to enhance the generalizability of LLMs. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 11:04:09 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 02:22:22 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Sun",
"Zhouhao",
""
],
[
"Kan",
"Zhiyuan",
""
],
[
"Ding",
"Xiao",
""
],
[
"Du",
"Li",
""
],
[
"Zhao",
"Yang",
""
],
[
"Qin",
"Bing",
""
],
[
"Liu",
"Ting",
""
]
] |
2505.16552 | Wenhui Tan | Wenhui Tan, Jiaze Li, Jianzhong Ju, Zhenbo Luo, Jian Luan, Ruihua Song | Think Silently, Think Fast: Dynamic Latent Compression of LLM Reasoning Chains | 15 pages, 8 figures | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large Language Models (LLMs) achieve superior performance through Chain-of-Thought (CoT) reasoning, but these token-level reasoning chains are computationally expensive and inefficient. In this paper, we introduce Compressed Latent Reasoning (CoLaR), a novel framework that dynamically compresses reasoning processes in latent space through a two-stage training approach. First, during supervised fine-tuning, CoLaR extends beyond next-token prediction by incorporating an auxiliary next compressed embedding prediction objective. This process merges embeddings of consecutive tokens using a compression factor randomly sampled from a predefined range, and trains a specialized latent head to predict distributions of subsequent compressed embeddings. Second, we enhance CoLaR through reinforcement learning (RL) that leverages the latent head's non-deterministic nature to explore diverse reasoning paths and exploit more compact ones. This approach enables CoLaR to: i) perform reasoning at a dense latent level (i.e., silently), substantially reducing reasoning chain length, and ii) dynamically adjust reasoning speed at inference time by simply prompting the desired compression factor. Extensive experiments across four mathematical reasoning datasets demonstrate that CoLaR achieves 14.1% higher accuracy than latent-based baseline methods at comparable compression ratios, and reduces reasoning chain length by 53.3% with only 4.8% performance degradation compared to explicit CoT method. Moreover, when applied to more challenging mathematical reasoning tasks, our RL-enhanced CoLaR demonstrates performance gains of up to 5.4% while dramatically reducing latent reasoning chain length by 82.8%. The code and models will be released upon acceptance. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 11:40:26 GMT"
},
{
"version": "v2",
"created": "Fri, 23 May 2025 05:02:47 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 05:41:16 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Tan",
"Wenhui",
""
],
[
"Li",
"Jiaze",
""
],
[
"Ju",
"Jianzhong",
""
],
[
"Luo",
"Zhenbo",
""
],
[
"Luan",
"Jian",
""
],
[
"Song",
"Ruihua",
""
]
] |
2505.16561 | Jannis Becktepe | Jannis Becktepe, Leona Hennig, Steffen Oeltze-Jafra, Marius Lindauer | Auto-nnU-Net: Towards Automated Medical Image Segmentation | 31 pages, 19 figures. Accepted for publication at AutoML 2025 | null | null | null | cs.CV cs.AI | http://creativecommons.org/licenses/by/4.0/ | Medical Image Segmentation (MIS) includes diverse tasks, from bone to organ segmentation, each with its own challenges in finding the best segmentation model. The state-of-the-art AutoML-related MIS-framework nnU-Net automates many aspects of model configuration but remains constrained by fixed hyperparameters and heuristic design choices. As a full-AutoML framework for MIS, we propose Auto-nnU-Net, a novel nnU-Net variant enabling hyperparameter optimization (HPO), neural architecture search (NAS), and hierarchical NAS (HNAS). Additionally, we propose Regularized PriorBand to balance model accuracy with the computational resources required for training, addressing the resource constraints often faced in real-world medical settings that limit the feasibility of extensive training procedures. We evaluate our approach across diverse MIS datasets from the well-established Medical Segmentation Decathlon, analyzing the impact of AutoML techniques on segmentation performance, computational efficiency, and model design choices. The results demonstrate that our AutoML approach substantially improves the segmentation performance of nnU-Net on 6 out of 10 datasets and is on par on the other datasets while maintaining practical resource requirements. Our code is available at https://github.com/automl/AutoNNUnet. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 11:52:16 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 06:52:37 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 06:46:06 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Becktepe",
"Jannis",
""
],
[
"Hennig",
"Leona",
""
],
[
"Oeltze-Jafra",
"Steffen",
""
],
[
"Lindauer",
"Marius",
""
]
] |
2505.16562 | Haihao Shi | Haihao Shi, Zhenyang Huang, Qiyu Yan, Jun Li, Guoliang L\"u, Xuefei Chen | A Machine Learning Pipeline for Hunting Hidden Axion Signals in Pulsar Dispersion Measurements | 9 pages, 7 figures, Updated the paper structure and references | null | null | null | astro-ph.HE | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In axion models, interactions between axions and electromagnetic waves induce frequency-dependent time delays determined by the axion mass and decay constant. These small delays are difficult to detect, limiting the effectiveness of traditional methods. We compute such delays under realistic radio telescope conditions and identify a prominent dispersive feature near half the axion mass, which appears non-divergent within the limits of observational resolution. Based on this, we develop a machine learning pipeline that achieves 95\% classification accuracy and demonstrates robust detection performance in low signal-to-noise regimes. When applied to PSR J1933-6211, no axion-induced delay is detected within current sensitivity limits. Future improvements in optical clock precision and telescope bandwidth, particularly with instruments such as the Qitai Radio Telescope, may enhance constraints on the axion decay constant by up to four orders of magnitude in the $10^{-6} \sim 10^{-4}$ eV mass range. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 11:52:56 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 08:58:16 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Shi",
"Haihao",
""
],
[
"Huang",
"Zhenyang",
""
],
[
"Yan",
"Qiyu",
""
],
[
"Li",
"Jun",
""
],
[
"Lü",
"Guoliang",
""
],
[
"Chen",
"Xuefei",
""
]
] |
2505.16662 | Chuan Huang | Chuan Huang, Gustaf Hendeby, Isaac Skog | Joint Magnetometer-IMU Calibration via Maximum A Posteriori Estimation | Fix a typo | null | null | null | cs.RO eess.SP | http://creativecommons.org/licenses/by/4.0/ | This paper presents a new approach for jointly calibrating magnetometers and inertial measurement units, focusing on improving calibration accuracy and computational efficiency. The proposed method formulates the calibration problem as a maximum a posteriori estimation problem, treating both the calibration parameters and orientation trajectory of the sensors as unknowns. This formulation enables efficient optimization with closed-form derivatives. The method is compared against two state-of-the-art approaches in terms of computational complexity and estimation accuracy. Simulation results demonstrate that the proposed method achieves lower root mean square error in calibration parameters while maintaining competitive computational efficiency. Further validation through real-world experiments confirms the practical benefits of our approach: it effectively reduces position drift in a magnetic field-aided inertial navigation system by more than a factor of two on most datasets. Moreover, the proposed method calibrated 30 magnetometers in less than 2 minutes. The contributions include a new calibration method, an analysis of existing methods, and a comprehensive empirical evaluation. Datasets and algorithms are made publicly available to promote reproducible research. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 13:27:42 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 15:29:31 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Huang",
"Chuan",
""
],
[
"Hendeby",
"Gustaf",
""
],
[
"Skog",
"Isaac",
""
]
] |
2505.16670 | Xiaobei Yan | Xiaobei Yan, Yiming Li, Zhaoxin Fan, Han Qiu, Tianwei Zhang | BitHydra: Towards Bit-flip Inference Cost Attack against Large Language Models | null | null | null | null | cs.CR cs.AI | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) have shown impressive capabilities across a wide range of applications, but their ever-increasing size and resource demands make them vulnerable to inference cost attacks, where attackers induce victim LLMs to generate the longest possible output content. In this paper, we revisit existing inference cost attacks and reveal that these methods can hardly produce large-scale malicious effects since they are self-targeting, where attackers are also the users and therefore have to execute attacks solely through the inputs, whose generated content will be charged by LLMs and can only directly influence themselves. Motivated by these findings, this paper introduces a new type of inference cost attacks (dubbed 'bit-flip inference cost attack') that target the victim model itself rather than its inputs. Specifically, we design a simple yet effective method (dubbed 'BitHydra') to effectively flip critical bits of model parameters. This process is guided by a loss function designed to suppress <EOS> token's probability with an efficient critical bit search algorithm, thus explicitly defining the attack objective and enabling effective optimization. We evaluate our method on 11 LLMs ranging from 1.5B to 14B parameters under both int8 and float16 settings. Experimental results demonstrate that with just 4 search samples and as few as 3 bit flips, BitHydra can force 100% of test prompts to reach the maximum generation length (e.g., 2048 tokens) on representative LLMs such as LLaMA3, highlighting its efficiency, scalability, and strong transferability across unseen inputs. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 13:36:00 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 10:55:21 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Yan",
"Xiaobei",
""
],
[
"Li",
"Yiming",
""
],
[
"Fan",
"Zhaoxin",
""
],
[
"Qiu",
"Han",
""
],
[
"Zhang",
"Tianwei",
""
]
] |
2505.16746 | Gervasi Herranz | A. S. Mi\~narro, G. Herranz | Emergent Orbital Dynamics in Strongly Spin-Orbit Coupled Systems | 21 pages, 8 figures | null | null | null | cond-mat.str-el | http://creativecommons.org/licenses/by/4.0/ | The interplay between spin and orbital degrees of freedom gives rise to a variety of emergent phases in correlated 4d and 5d transition-metal systems. Strong spin-orbit coupling (SOC) significantly alters Jahn-Teller (JT) physics, often suppressing static distortions or promoting dynamic fluctuations, thereby reducing or even quenching orbital polarization. While intersite hybridization is a fundamental aspect of crystalline solids, its role in shaping the dynamics of spin-orbit-entangled states has received comparatively little attention. Here, we show that electronic hopping can locally restore orbital polarization when the ground state is perturbed, even in the absence of static orbital order. Using a Matsubara lattice formalism, we analyze how local orbital perturbations propagate through correlated, spin-orbit-entangled systems. When intersite hopping is included, such perturbations induce short-range orbital polarization with a characteristic orthogonal response at nearest-neighbor sites. Although the energy scale of these hybridization-driven orbital reconstructions likely makes their detection challenging, they may still influence low-energy spectral features and interact with other excitations. These results underscore the importance of including orbital dynamics in the interpretation of spectroscopic data and provide a framework for understanding dynamical responses in spin-orbit-entangled materials. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 14:56:49 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 13:14:17 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Miñarro",
"A. S.",
""
],
[
"Herranz",
"G.",
""
]
] |
2505.16750 | Jonathan Ron E | Jonathan E. Ron and Ram M. Adar | Theory of cell size regulation during migration in adhered cells | 10 pages, 5 figures | null | null | null | physics.bio-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Cell migration is closely linked to cell shape, yet cell size is often assumed to remain constant. This assumption is challenged by recent experiments showing that cells undergo volume loss during spreading and swelling upon activation, with migration velocity correlated to cell size. In this Letter, we present a minimal theoretical framework for cellular size regulation and its influence on migration velocity. We connect cell size to membrane potential and active, actin-driven forces. Spatial inhomogeneities in these forces establish cell polarization and drive migration. Crucially, inhomogeneity is easier to establish over larger sizes, giving rise to a critical contact area, above which migration is possible. Our theory captures the coupled dynamics of cell volume, surface area, and motility and explains recent experiments on neutrophils. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 14:58:10 GMT"
},
{
"version": "v2",
"created": "Fri, 23 May 2025 04:11:11 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 14:15:45 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Ron",
"Jonathan E.",
""
],
[
"Adar",
"Ram M.",
""
]
] |
2505.16787 | Ashish Sundar | Ashish Sundar, Chunbo Luo, Xiaoyang Wang | Enter the Void - Planning to Seek Entropy When Reward is Scarce | 9 pages without appendix, 15 Figures, preprint | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Model-based reinforcement learning (MBRL) offers an intuitive way to increase the sample efficiency of model-free RL methods by simultaneously training a world model that learns to predict the future. MBRL methods have progressed by largely prioritising the actor; optimising the world model learning has been neglected meanwhile. Improving the fidelity of the world model and reducing its time to convergence can yield significant downstream benefits, one of which is improving the ensuing performance of any actor it may train. We propose a novel approach that anticipates and actively seeks out high-entropy states using short-horizon latent predictions generated by the world model, offering a principled alternative to traditional curiosity-driven methods that chase once-novel states well after they were stumbled into. While many model predictive control (MPC) based methods offer similar alternatives, they typically lack commitment, synthesising multi step plans after every step. To mitigate this, we present a hierarchical planner that dynamically decides when to replan, planning horizon length, and the weighting between reward and entropy. While our method can theoretically be applied to any model that trains its own actors with solely model generated data, we have applied it to just Dreamer as a proof of concept. Our method finishes the Miniworld procedurally generated mazes 50% faster than base Dreamer at convergence and the policy trained in imagination converges in only 60% of the environment steps that base Dreamer needs. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 15:28:50 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 12:28:55 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Sundar",
"Ashish",
""
],
[
"Luo",
"Chunbo",
""
],
[
"Wang",
"Xiaoyang",
""
]
] |
2505.16814 | Sowmya Vajjala | Gaurav Kamath and Sowmya Vajjala | Does Synthetic Data Help Named Entity Recognition for Low-Resource Languages? | pre-print | null | null | null | cs.CL | http://creativecommons.org/licenses/by-sa/4.0/ | Named Entity Recognition(NER) for low-resource languages aims to produce robust systems for languages where there is limited labeled training data available, and has been an area of increasing interest within NLP. Data augmentation for increasing the amount of low-resource labeled data is a common practice. In this paper, we explore the role of synthetic data in the context of multilingual, low-resource NER, considering 11 languages from diverse language families. Our results suggest that synthetic data does in fact hold promise for low-resource language NER, though we see significant variation between languages. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 15:50:47 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 18:32:38 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Kamath",
"Gaurav",
""
],
[
"Vajjala",
"Sowmya",
""
]
] |
2505.16892 | Luzhe Sun | Luzhe Sun, Jingtian Ji, Xiangshan Tan, Matthew R. Walter | FlashBack: Consistency Model-Accelerated Shared Autonomy | null | null | null | null | cs.RO | http://creativecommons.org/licenses/by/4.0/ | Shared autonomy is an enabling technology that provides users with control authority over robots that would otherwise be difficult if not impossible to directly control. Yet, standard methods make assumptions that limit their adoption in practice-for example, prior knowledge of the user's goals or the objective (i.e., reward) function that they wish to optimize, knowledge of the user's policy, or query-level access to the user during training. Diffusion-based approaches to shared autonomy do not make such assumptions and instead only require access to demonstrations of desired behaviors, while allowing the user to maintain control authority. However, these advantages have come at the expense of high computational complexity, which has made real-time shared autonomy all but impossible. To overcome this limitation, we propose Consistency Shared Autonomy (CSA), a shared autonomy framework that employs a consistency model-based formulation of diffusion. Key to CSA is that it employs the distilled probability flow of ordinary differential equations (PF ODE) to generate high-fidelity samples in a single step. This results in inference speeds significantly than what is possible with previous diffusion-based approaches to shared autonomy, enabling real-time assistance in complex domains with only a single function evaluation. Further, by intervening on flawed actions at intermediate states of the PF ODE, CSA enables varying levels of assistance. We evaluate CSA on a variety of challenging simulated and real-world robot control problems, demonstrating significant improvements over state-of-the-art methods both in terms of task performance and computational efficiency. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 16:50:53 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:58:28 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Sun",
"Luzhe",
""
],
[
"Ji",
"Jingtian",
""
],
[
"Tan",
"Xiangshan",
""
],
[
"Walter",
"Matthew R.",
""
]
] |
2505.16900 | Jintian Shao | Jintian Shao, Yiming Cheng, Hongyi Huang, Jiayi Wu, Beiwen Zhang, Zhiyu Wu, You Shan, Mingkai Zheng | Power-Law Decay Loss for Large Language Model Finetuning: Focusing on Information Sparsity to Enhance Generation Quality | We are withdrawing this submission as the underlying experiment is currently incomplete. We require additional time to gather more data and supplement the existing findings to ensure a comprehensive and robust presentation. We intend to resubmit once these additions are finalized | null | null | null | cs.CL cs.LG | http://creativecommons.org/licenses/by/4.0/ | During the finetuning stage of text generation tasks, standard cross-entropy loss treats all tokens equally. This can lead models to overemphasize high-frequency, low-information tokens, neglecting lower-frequency tokens crucial for specificity and informativeness in generated content. This paper introduces a novel loss function, Power-Law Decay Loss (PDL), specifically designed to optimize the finetuning process for text generation. The core motivation for PDL stems from observations in information theory and linguistics: the informativeness of a token is often inversely proportional to its frequency of occurrence. PDL re-weights the contribution of each token in the standard cross-entropy loss based on its frequency in the training corpus, following a power-law decay. Specifically, the weights for high-frequency tokens are reduced, while low-frequency, information-dense tokens are assigned higher weights. This mechanism guides the model during finetuning to focus more on learning and generating tokens that convey specific and unique information, thereby enhancing the quality, diversity, and informativeness of the generated text. We theoretically elaborate on the motivation and construction of PDL and discuss its potential applications and advantages across various text generation finetuning tasks, such as abstractive summarization, dialogue systems, and style transfer. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 16:59:26 GMT"
},
{
"version": "v2",
"created": "Fri, 23 May 2025 11:27:02 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 08:30:32 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Shao",
"Jintian",
""
],
[
"Cheng",
"Yiming",
""
],
[
"Huang",
"Hongyi",
""
],
[
"Wu",
"Jiayi",
""
],
[
"Zhang",
"Beiwen",
""
],
[
"Wu",
"Zhiyu",
""
],
[
"Shan",
"You",
""
],
[
"Zheng",
"Mingkai",
""
]
] |
2505.17013 | Kevin Lu | Kevin Lu, Nicky Kriplani, Rohit Gandikota, Minh Pham, David Bau, Chinmay Hegde, Niv Cohen | When Are Concepts Erased From Diffusion Models? | Project Page: https://nyu-dice-lab.github.io/when-are-concepts-erased/ | null | null | null | cs.LG cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Concept erasure, the ability to selectively prevent a model from generating specific concepts, has attracted growing interest, with various approaches emerging to address the challenge. However, it remains unclear how thoroughly these methods erase the target concept. We begin by proposing two conceptual models for the erasure mechanism in diffusion models: (i) reducing the likelihood of generating the target concept, and (ii) interfering with the model's internal guidance mechanisms. To thoroughly assess whether a concept has been truly erased from the model, we introduce a suite of independent evaluations. Our evaluation framework includes adversarial attacks, novel probing techniques, and analysis of the model's alternative generations in place of the erased concept. Our results shed light on the tension between minimizing side effects and maintaining robustness to adversarial prompts. Broadly, our work underlines the importance of comprehensive evaluation for erasure in diffusion models. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 17:59:09 GMT"
},
{
"version": "v2",
"created": "Fri, 23 May 2025 20:12:20 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 17:43:05 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Lu",
"Kevin",
""
],
[
"Kriplani",
"Nicky",
""
],
[
"Gandikota",
"Rohit",
""
],
[
"Pham",
"Minh",
""
],
[
"Bau",
"David",
""
],
[
"Hegde",
"Chinmay",
""
],
[
"Cohen",
"Niv",
""
]
] |
2505.17049 | David Rozado | David Rozado | Gender and Positional Biases in LLM-Based Hiring Decisions: Evidence from Comparative CV/R\'esum\'e Evaluations | null | null | null | null | cs.CL cs.AI cs.CY | http://creativecommons.org/licenses/by-nc-sa/4.0/ | This study examines the behavior of Large Language Models (LLMs) when evaluating professional candidates based on their resumes or curricula vitae (CVs). In an experiment involving 22 leading LLMs, each model was systematically given one job description along with a pair of profession-matched CVs, one bearing a male first name, the other a female first name, and asked to select the more suitable candidate for the job. Each CV pair was presented twice, with names swapped to ensure that any observed preferences in candidate selection stemmed from gendered names cues. Despite identical professional qualifications across genders, all LLMs consistently favored female-named candidates across 70 different professions. Adding an explicit gender field (male/female) to the CVs further increased the preference for female applicants. When gendered names were replaced with gender-neutral identifiers "Candidate A" and "Candidate B", several models displayed a preference to select "Candidate A". Counterbalancing gender assignment between these gender-neutral identifiers resulted in gender parity in candidate selection. When asked to rate CVs in isolation rather than compare pairs, LLMs assigned slightly higher average scores to female CVs overall, but the effect size was negligible. Including preferred pronouns (he/him or she/her) next to a candidate's name slightly increased the odds of the candidate being selected regardless of gender. Finally, most models exhibited a substantial positional bias to select the candidate listed first in the prompt. These findings underscore the need for caution when deploying LLMs in high-stakes autonomous decision-making contexts and raise doubts about whether LLMs consistently apply principled reasoning. | [
{
"version": "v1",
"created": "Fri, 16 May 2025 06:19:35 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 00:07:04 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Rozado",
"David",
""
]
] |
2505.17117 | Chen Shani | Chen Shani, Dan Jurafsky, Yann LeCun, Ravid Shwartz-Ziv | From Tokens to Thoughts: How LLMs and Humans Trade Compression for Meaning | null | null | null | null | cs.CL cs.AI cs.IT math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Humans organize knowledge into compact categories through semantic compression by mapping diverse instances to abstract representations while preserving meaning (e.g., robin and blue jay are both birds; most birds can fly). These concepts reflect a trade-off between expressive fidelity and representational simplicity. Large Language Models (LLMs) demonstrate remarkable linguistic abilities, yet whether their internal representations strike a human-like trade-off between compression and semantic fidelity is unclear. We introduce a novel information-theoretic framework, drawing from Rate-Distortion Theory and the Information Bottleneck principle, to quantitatively compare these strategies. Analyzing token embeddings from a diverse suite of LLMs against seminal human categorization benchmarks, we uncover key divergences. While LLMs form broad conceptual categories that align with human judgment, they struggle to capture the fine-grained semantic distinctions crucial for human understanding. More fundamentally, LLMs demonstrate a strong bias towards aggressive statistical compression, whereas human conceptual systems appear to prioritize adaptive nuance and contextual richness, even if this results in lower compressional efficiency by our measures. These findings illuminate critical differences between current AI and human cognitive architectures, guiding pathways toward LLMs with more human-aligned conceptual representations. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 16:29:00 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 21:13:36 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Shani",
"Chen",
""
],
[
"Jurafsky",
"Dan",
""
],
[
"LeCun",
"Yann",
""
],
[
"Shwartz-Ziv",
"Ravid",
""
]
] |
2505.17192 | Zongmeng Yang | Zongmeng Yang, Xingyue Yang, Jianhua Wang, Rui Peng, Lee Ching Hua, Lay Kee Ang, Jing Lu, Yee Sin Ang, and Shibo Fang | Unconventional tunnel magnetoresistance scaling with altermagnets | null | null | null | null | cond-mat.mes-hall cond-mat.mtrl-sci physics.comp-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In conventional magnetic tunnel junctions (MTJs), the tunnel magnetoresistance (TMR) typically increases with barrier thickness as electron transmission in the antiparallel configuration decays faster than that of the parallel configuration. In this work, we reveal an anomalous scaling effect in altermagnetic tunnel junctions (AMTJs), where the TMR decreases anomalously with an increasing barrier thickness. The anomalous scaling originates from the overlapping spin-split branches form a transmission path that cannot be suppressed in the antiparallel state. Such phenomena is explained by adouble-barrier model and is further demonstrated using ab initio quantum transport simulations in 2D V2Te2O/Cr2Se2O/V2Te2O-based AMTJ, where the TMR anomalously decreases from 220% to 40% as the layer number of Cr2Se2O increases from 1 to 5. Our work identifies a peculiar unexpected transport characteristic of AMTJ, providing a fundamental limit on AMTJ device design and illustrating the potential optimal design of AMTJ at the ultrascaled monolayer limit. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 18:02:37 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Yang",
"Zongmeng",
""
],
[
"Yang",
"Xingyue",
""
],
[
"Wang",
"Jianhua",
""
],
[
"Peng",
"Rui",
""
],
[
"Hua",
"Lee Ching",
""
],
[
"Ang",
"Lay Kee",
""
],
[
"Lu",
"Jing",
""
],
[
"Ang",
"Yee Sin",
""
],
[
"Fang",
"Shibo",
""
]
] |
2505.17262 | Rodrigo Holanda | L. R. Cola\c{c}o, R. F. L. Holanda, Z. C. Santana, R. Silva | A Joint Analysis of Strong Lensing and Type Ia Supernovae to Determine the Hubble Constant | 10 pages, 3 figures, I table, Version to appear in The European Physical Journal C | Eur. Phys. J. C (2025) 85:577 | 10.1140/epjc/s10052-025-14315-6 | null | astro-ph.CO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present a cosmological model-independent determination of the Hubble constant, $H_0$, by combining time-delay measurements from seven TDCOSMO systems, Einstein radius measurements, and Type Ia Supernovae data sourced from the Pantheon+ sample. For each lens of time-delay system, we calculate the angular diameter distance $D_{A_l}$ using the product $D^{\textrm{Obs}}(z_l) \cdot D_{A,\Delta t}^{\textrm{Obs}}(z_l, z_s)$, where $D^{\textrm{Obs}}(z_l)$ is reconstructed via Gaussian Processes from 99 Einstein radius measurements, and $D_{A,\Delta t}^{\textrm{Obs}}(z_l,z_s)$ is the time-delay angular distance. We also reconstruct the unanchored luminosity distance $H_0 D_L(z_l)$ from supernova data. By using the cosmic distance duality relation validity, we anchor $D_{A_l}$ and $H_0 D_L(z_l)$ to infer $H_0 = 70.55 \pm 7.44$ km/s/Mpc (68\% CL). Our result, though not resolving the Hubble tension, offers a cosmological model-independent consistency check and highlights the potential of using strong lensing and supernovae data via the cosmic distance duality relation to constrain $H_0$. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 20:16:33 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Colaço",
"L. R.",
""
],
[
"Holanda",
"R. F. L.",
""
],
[
"Santana",
"Z. C.",
""
],
[
"Silva",
"R.",
""
]
] |
2505.17266 | Cehao Yang | Cehao Yang, Xueyuan Lin, Chengjin Xu, Xuhui Jiang, Xiaojun Wu, Honghao Liu, Hui Xiong, Jian Guo | Select2Reason: Efficient Instruction-Tuning Data Selection for Long-CoT Reasoning | null | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by/4.0/ | A practical approach to activate long chain-of-thoughts reasoning ability in pre-trained large language models is to perform supervised fine-tuning on instruction datasets synthesized by strong Large Reasoning Models such as DeepSeek-R1, offering a cost-effective alternative to reinforcement learning. However, large-scale instruction sets with more than 100k samples incur significant training overhead, while effective strategies for automatic long-CoT instruction selection still remain unexplored. In this work, we propose Select2Reason, a novel and efficient instruction-tuning data selection framework for long-CoT reasoning. From the perspective of emergence of rethinking behaviors like self-correction and backtracking, we investigate common metrics that may determine the quality of long-CoT reasoning instructions. Select2Reason leverages a quantifier to estimate difficulty of question and jointly incorporates a reasoning trace length-based heuristic through a weighted scheme for ranking to prioritize high-utility examples. Empirical results on OpenR1-Math-220k demonstrate that fine-tuning LLM on only 10% of the data selected by Select2Reason achieves performance competitive with or superior to full-data tuning and open-source baseline OpenR1-Qwen-7B across three competition-level and six comprehensive mathematical benchmarks. Further experiments highlight the scalability in varying data size, efficiency during inference, and its adaptability to other instruction pools with minimal cost. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 20:24:08 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 15:50:50 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Yang",
"Cehao",
""
],
[
"Lin",
"Xueyuan",
""
],
[
"Xu",
"Chengjin",
""
],
[
"Jiang",
"Xuhui",
""
],
[
"Wu",
"Xiaojun",
""
],
[
"Liu",
"Honghao",
""
],
[
"Xiong",
"Hui",
""
],
[
"Guo",
"Jian",
""
]
] |
2505.17278 | C\'icero Zanette | Elisa G. Vergamini (1), Leonardo F. Dos Santos (1), C\'icero Zanette (1), Yecid Moreno (1), Felix M. Escalante (2), Thiago Boaventura (1) ((1) S\~ao Carlos School of Engineering - University of S\~ao Paulo, (2) Institute of Science and Technology Sorocaba - S\~ao Paulo State University) | Construction of an Impedance Control Test Bench | 10 pages, 23 figures | null | 10.26678/ABCM.COBEM2023.COB2023-0847 | null | cs.RO cs.SY eess.SY | http://creativecommons.org/licenses/by/4.0/ | Controlling the physical interaction with the environment or objects, as humans do, is a shared requirement across different types of robots. To effectively control this interaction, it is necessary to control the power delivered to the load, that is, the interaction force and the interaction velocity. However, it is not possible to control these two quantities independently at the same time. An alternative is to control the relation between them, with Impedance and Admittance control, for example. The Impedance Control 2 Dimensions (IC2D) bench is a test bench designed to allow the performance analysis of different actuators and controllers at the joint level. Therefore, it was designed to be as versatile as possible, to allow the combination of linear and/or rotational motions, to use electric and/or hydraulic actuators, with loads known and defined by the user. The bench adheres to a set of requirements defined by the demands of the research group, to be a reliable, backlash-free mechatronic system to validate system dynamics models and controller designs, as well as a valuable experimental setup for benchmarking electric and hydraulic actuators. This article presents the mechanical, electrical, and hydraulic configurations used to ensure the robustness and reliability of the test bench. Benches similar to this one are commonly found in robotics laboratories around the world. However, the IC2D stands out for its versatility and reliability, as well as for supporting hydraulic and electric actuators. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 20:53:35 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Vergamini",
"Elisa G.",
"",
"São Carlos School of Engineering - University of São Paulo"
],
[
"Santos",
"Leonardo F. Dos",
"",
"São Carlos School of Engineering - University of São Paulo"
],
[
"Zanette",
"Cícero",
"",
"São Carlos School of Engineering - University of São Paulo"
],
[
"Moreno",
"Yecid",
"",
"São Carlos School of Engineering - University of São Paulo"
],
[
"Escalante",
"Felix M.",
"",
"Institute of Science and Technology Sorocaba - São Paulo State University"
],
[
"Boaventura",
"Thiago",
"",
"São Carlos School of Engineering - University of São Paulo"
]
] |
2505.17370 | Qilin Wang | Qilin Wang | FRIREN: Beyond Trajectories -- A Spectral Lens on Time | 37 pages, 4 figures. Submitted to NeurIPS 2025. Public code at https://anonymous.4open.science/r/LTSF_model-C6B8/ | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Long-term time-series forecasting (LTSF) models are often presented as general-purpose solutions that can be applied across domains, implicitly assuming that all data is pointwise predictable. Using chaotic systems such as Lorenz-63 as a case study, we argue that geometric structure - not pointwise prediction - is the right abstraction for a dynamic-agnostic foundational model. Minimizing the Wasserstein-2 distance (W2), which captures geometric changes, and providing a spectral view of dynamics are essential for long-horizon forecasting. Our model, FRIREN (Flow-inspired Representations via Interpretable Eigen-networks), implements an augmented normalizing-flow block that embeds data into a normally distributed latent representation. It then generates a W2-efficient optimal path that can be decomposed into rotation, scaling, inverse rotation, and translation. This architecture yields locally generated, geometry-preserving predictions that are independent of the underlying dynamics, and a global spectral representation that functions as a finite Koopman operator with a small modification. This enables practitioners to identify which modes grow, decay, or oscillate, both locally and system-wide. FRIREN achieves an MSE of 11.4, MAE of 1.6, and SWD of 0.96 on Lorenz-63 in a 336-in, 336-out, dt=0.01 setting, surpassing TimeMixer (MSE 27.3, MAE 2.8, SWD 2.1). The model maintains effective prediction for 274 out of 336 steps, approximately 2.5 Lyapunov times. On Rossler (96-in, 336-out), FRIREN achieves an MSE of 0.0349, MAE of 0.0953, and SWD of 0.0170, outperforming TimeMixer's MSE of 4.3988, MAE of 0.886, and SWD of 3.2065. FRIREN is also competitive on standard LTSF datasets such as ETT and Weather. By connecting modern generative flows with classical spectral analysis, FRIREN makes long-term forecasting both accurate and interpretable, setting a new benchmark for LTSF model design. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 00:52:13 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:16:57 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wang",
"Qilin",
""
]
] |
2505.17589 | Zhihao Du | Zhihao Du, Changfeng Gao, Yuxuan Wang, Fan Yu, Tianyu Zhao, Hao Wang, Xiang Lv, Hui Wang, Chongjia Ni, Xian Shi, Keyu An, Guanrou Yang, Yabin Li, Yanni Chen, Zhifu Gao, Qian Chen, Yue Gu, Mengzhe Chen, Yafeng Chen, Shiliang Zhang, Wen Wang, Jieping Ye | CosyVoice 3: Towards In-the-wild Speech Generation via Scaling-up and Post-training | Preprint, work in progress | null | null | null | cs.SD cs.AI eess.AS | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In our prior works, we introduced a scalable streaming speech synthesis model, CosyVoice 2, which integrates a large language model (LLM) and a chunk-aware flow matching (FM) model, and achieves low-latency bi-streaming speech synthesis and human-parity quality. Despite these advancements, CosyVoice 2 exhibits limitations in language coverage, domain diversity, data volume, text formats, and post-training techniques. In this paper, we present CosyVoice 3, an improved model designed for zero-shot multilingual speech synthesis in the wild, surpassing its predecessor in content consistency, speaker similarity, and prosody naturalness. Key features of CosyVoice 3 include: 1) A novel speech tokenizer to improve prosody naturalness, developed via supervised multi-task training, including automatic speech recognition, speech emotion recognition, language identification, audio event detection, and speaker analysis. 2) A new differentiable reward model for post-training applicable not only to CosyVoice 3 but also to other LLM-based speech synthesis models. 3) Dataset Size Scaling: Training data is expanded from ten thousand hours to one million hours, encompassing 9 languages and 18 Chinese dialects across various domains and text formats. 4) Model Size Scaling: Model parameters are increased from 0.5 billion to 1.5 billion, resulting in enhanced performance on our multilingual benchmark due to the larger model capacity. These advancements contribute significantly to the progress of speech synthesis in the wild. We encourage readers to listen to the demo at https://funaudiollm.github.io/cosyvoice3. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 07:55:21 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:48:34 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Du",
"Zhihao",
""
],
[
"Gao",
"Changfeng",
""
],
[
"Wang",
"Yuxuan",
""
],
[
"Yu",
"Fan",
""
],
[
"Zhao",
"Tianyu",
""
],
[
"Wang",
"Hao",
""
],
[
"Lv",
"Xiang",
""
],
[
"Wang",
"Hui",
""
],
[
"Ni",
"Chongjia",
""
],
[
"Shi",
"Xian",
""
],
[
"An",
"Keyu",
""
],
[
"Yang",
"Guanrou",
""
],
[
"Li",
"Yabin",
""
],
[
"Chen",
"Yanni",
""
],
[
"Gao",
"Zhifu",
""
],
[
"Chen",
"Qian",
""
],
[
"Gu",
"Yue",
""
],
[
"Chen",
"Mengzhe",
""
],
[
"Chen",
"Yafeng",
""
],
[
"Zhang",
"Shiliang",
""
],
[
"Wang",
"Wen",
""
],
[
"Ye",
"Jieping",
""
]
] |
2505.17595 | Li Lin | Li Lin, Xinyu Hu, Xiaojun Wan | NeUQI: Near-Optimal Uniform Quantization Parameter Initialization | 9 pages, under review | null | null | null | cs.LG cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) achieve impressive performance across domains but face significant challenges when deployed on consumer-grade GPUs or personal devices such as laptops, due to high memory consumption and inference costs. Post-training quantization (PTQ) of LLMs offers a promising solution that reduces their memory footprint and decoding latency. In practice, PTQ with uniform quantization representation is favored for its efficiency and ease of deployment since uniform quantization is widely supported by mainstream hardware and software libraries. Recent studies on $\geq 2$-bit uniform quantization have led to noticeable improvements in post-quantization model performance; however, they primarily focus on quantization methodologies, while the initialization of quantization parameters is underexplored and still relies on the suboptimal Min-Max strategies. In this work, we propose NeUQI, a method devoted to efficiently determining near-optimal initial parameters for uniform quantization. NeUQI is orthogonal to prior quantization methodologies and can seamlessly integrate with them. The experiments with the LLaMA and Qwen families on various tasks demonstrate that our NeUQI consistently outperforms existing methods. Furthermore, when combined with a lightweight distillation strategy, NeUQI can achieve superior performance to PV-tuning, a much more resource-intensive approach. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 07:59:46 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 04:25:03 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Lin",
"Li",
""
],
[
"Hu",
"Xinyu",
""
],
[
"Wan",
"Xiaojun",
""
]
] |
2505.17621 | Jingtong Gao | Jingtong Gao, Ling Pan, Yejing Wang, Rui Zhong, Chi Lu, Qingpeng Cai, Peng Jiang, Xiangyu Zhao | Navigate the Unknown: Enhancing LLM Reasoning with Intrinsic Motivation Guided Exploration | null | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Reinforcement learning (RL) has emerged as a pivotal method for improving the reasoning capabilities of Large Language Models (LLMs). However, prevalent RL approaches such as Proximal Policy Optimization (PPO) and Group-Regularized Policy Optimization (GRPO) face critical limitations due to their reliance on sparse outcome-based rewards and inadequate mechanisms for incentivizing exploration. These limitations result in inefficient guidance for multi-step reasoning processes. Specifically, sparse reward signals fail to deliver effective or sufficient feedback, particularly for challenging problems. Furthermore, such reward structures induce systematic biases that prioritize exploitation of familiar trajectories over novel solution discovery. These shortcomings critically hinder performance in complex reasoning tasks, which inherently demand iterative refinement across ipntermediate steps. To address these challenges, we propose an Intrinsic Motivation guidEd exploratioN meThOd foR LLM Reasoning (i-MENTOR), a novel method designed to both deliver dense rewards and amplify explorations in the RL-based training paradigm. i-MENTOR introduces three key innovations: trajectory-aware exploration rewards that mitigate bias in token-level strategies while maintaining computational efficiency; dynamic reward scaling to stabilize exploration and exploitation in large action spaces; and advantage-preserving reward implementation that maintains advantage distribution integrity while incorporating exploratory guidance. Experiments across three public datasets demonstrate i-MENTOR's effectiveness with a 22.39% improvement on the difficult dataset Countdown-4. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 08:30:28 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 04:06:17 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Gao",
"Jingtong",
""
],
[
"Pan",
"Ling",
""
],
[
"Wang",
"Yejing",
""
],
[
"Zhong",
"Rui",
""
],
[
"Lu",
"Chi",
""
],
[
"Cai",
"Qingpeng",
""
],
[
"Jiang",
"Peng",
""
],
[
"Zhao",
"Xiangyu",
""
]
] |
2505.17629 | Yuheng Lu | Yuheng Lu, Qian Yu, Hongru Wang, Zeming Liu, Wei Su, Yanping Liu, Yuhang Guo, Maocheng Liang, Yunhong Wang, Haifeng Wang | TransBench: Breaking Barriers for Transferable Graphical User Interface Agents in Dynamic Digital Environments | Accepted by ACL 2025 Findings | null | null | null | cs.HC cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Graphical User Interface (GUI) agents, which autonomously operate on digital interfaces through natural language instructions, hold transformative potential for accessibility, automation, and user experience. A critical aspect of their functionality is grounding - the ability to map linguistic intents to visual and structural interface elements. However, existing GUI agents often struggle to adapt to the dynamic and interconnected nature of real-world digital environments, where tasks frequently span multiple platforms and applications while also being impacted by version updates. To address this, we introduce TransBench, the first benchmark designed to systematically evaluate and enhance the transferability of GUI agents across three key dimensions: cross-version transferability (adapting to version updates), cross-platform transferability (generalizing across platforms like iOS, Android, and Web), and cross-application transferability (handling tasks spanning functionally distinct apps). TransBench includes 15 app categories with diverse functionalities, capturing essential pages across versions and platforms to enable robust evaluation. Our experiments demonstrate significant improvements in grounding accuracy, showcasing the practical utility of GUI agents in dynamic, real-world environments. Our code and data will be publicly available at GitHub. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 08:39:06 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:41:51 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Lu",
"Yuheng",
""
],
[
"Yu",
"Qian",
""
],
[
"Wang",
"Hongru",
""
],
[
"Liu",
"Zeming",
""
],
[
"Su",
"Wei",
""
],
[
"Liu",
"Yanping",
""
],
[
"Guo",
"Yuhang",
""
],
[
"Liang",
"Maocheng",
""
],
[
"Wang",
"Yunhong",
""
],
[
"Wang",
"Haifeng",
""
]
] |
2505.17659 | Xiaolong Tang | Xiaolong Tang, Meina Kan, Shiguang Shan, Xilin Chen | Plan-R1: Safe and Feasible Trajectory Planning as Language Modeling | null | null | null | null | cs.RO cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Safe and feasible trajectory planning is essential for real-world autonomous driving systems. However, existing learning-based planning methods often rely on expert demonstrations, which not only lack explicit safety awareness but also risk inheriting unsafe behaviors such as speeding from suboptimal human driving data. Inspired by the success of large language models, we propose Plan-R1, a novel two-stage trajectory planning framework that formulates trajectory planning as a sequential prediction task, guided by explicit planning principles such as safety, comfort, and traffic rule compliance. In the first stage, we train an autoregressive trajectory predictor via next motion token prediction on expert data. In the second stage, we design rule-based rewards (e.g., collision avoidance, speed limits) and fine-tune the model using Group Relative Policy Optimization (GRPO), a reinforcement learning strategy, to align its predictions with these planning principles. Experiments on the nuPlan benchmark demonstrate that our Plan-R1 significantly improves planning safety and feasibility, achieving state-of-the-art performance. Our code will be made public soon. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 09:22:19 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:51:12 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Tang",
"Xiaolong",
""
],
[
"Kan",
"Meina",
""
],
[
"Shan",
"Shiguang",
""
],
[
"Chen",
"Xilin",
""
]
] |
2505.17667 | Fanqi Wan | Fanqi Wan, Weizhou Shen, Shengyi Liao, Yingcheng Shi, Chenliang Li, Ziyi Yang, Ji Zhang, Fei Huang, Jingren Zhou, Ming Yan | QwenLong-L1: Towards Long-Context Large Reasoning Models with Reinforcement Learning | Technical Report | null | null | null | cs.CL | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent large reasoning models (LRMs) have demonstrated strong reasoning capabilities through reinforcement learning (RL). These improvements have primarily been observed within the short-context reasoning tasks. In contrast, extending LRMs to effectively process and reason on long-context inputs via RL remains a critical unsolved challenge. To bridge this gap, we first formalize the paradigm of long-context reasoning RL, and identify key challenges in suboptimal training efficiency and unstable optimization process. To address these issues, we propose QwenLong-L1, a framework that adapts short-context LRMs to long-context scenarios via progressive context scaling. Specifically, we utilize a warm-up supervised fine-tuning (SFT) stage to establish a robust initial policy, followed by a curriculum-guided phased RL technique to stabilize the policy evolution, and enhanced with a difficulty-aware retrospective sampling strategy to incentivize the policy exploration. Experiments on seven long-context document question-answering benchmarks demonstrate that QwenLong-L1-32B outperforms flagship LRMs like OpenAI-o3-mini and Qwen3-235B-A22B, achieving performance on par with Claude-3.7-Sonnet-Thinking, demonstrating leading performance among state-of-the-art LRMs. This work advances the development of practical long-context LRMs capable of robust reasoning across information-intensive environments. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 09:31:55 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 09:39:47 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Wan",
"Fanqi",
""
],
[
"Shen",
"Weizhou",
""
],
[
"Liao",
"Shengyi",
""
],
[
"Shi",
"Yingcheng",
""
],
[
"Li",
"Chenliang",
""
],
[
"Yang",
"Ziyi",
""
],
[
"Zhang",
"Ji",
""
],
[
"Huang",
"Fei",
""
],
[
"Zhou",
"Jingren",
""
],
[
"Yan",
"Ming",
""
]
] |
2505.17708 | Dingling Yao | Dingling Yao, Shimeng Huang, Riccardo Cadei, Kun Zhang, Francesco Locatello | The Third Pillar of Causal Analysis? A Measurement Perspective on Causal Representations | 22 pages, 12 figures, 2 tables | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Causal reasoning and discovery, two fundamental tasks of causal analysis, often face challenges in applications due to the complexity, noisiness, and high-dimensionality of real-world data. Despite recent progress in identifying latent causal structures using causal representation learning (CRL), what makes learned representations useful for causal downstream tasks and how to evaluate them are still not well understood. In this paper, we reinterpret CRL using a measurement model framework, where the learned representations are viewed as proxy measurements of the latent causal variables. Our approach clarifies the conditions under which learned representations support downstream causal reasoning and provides a principled basis for quantitatively assessing the quality of representations using a new Test-based Measurement EXclusivity (T-MEX) score. We validate T-MEX across diverse causal inference scenarios, including numerical simulations and real-world ecological video analysis, demonstrating that the proposed framework and corresponding score effectively assess the identification of learned representations and their usefulness for causal downstream tasks. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 10:25:17 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:51:53 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Yao",
"Dingling",
""
],
[
"Huang",
"Shimeng",
""
],
[
"Cadei",
"Riccardo",
""
],
[
"Zhang",
"Kun",
""
],
[
"Locatello",
"Francesco",
""
]
] |
2505.17854 | Lukas Koller | Lukas Koller, Tobias Ladner, Matthias Althoff | Out of the Shadows: Exploring a Latent Space for Neural Network Verification | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Neural networks are ubiquitous. However, they are often sensitive to small input changes. Hence, to prevent unexpected behavior in safety-critical applications, their formal verification -- a notoriously hard problem -- is necessary. Many state-of-the-art verification algorithms use reachability analysis or abstract interpretation to enclose the set of possible outputs of a neural network. Often, the verification is inconclusive due to the conservatism of the enclosure. To address this problem, we design a novel latent space for formal verification that enables the transfer of output specifications to the input space for an iterative specification-driven input refinement, i.e., we iteratively reduce the set of possible inputs to only enclose the unsafe ones. The latent space is constructed from a novel view of projection-based set representations, e.g., zonotopes, which are commonly used in reachability analysis of neural networks. A projection-based set representation is a "shadow" of a higher-dimensional set -- a latent space -- that does not change during a set propagation through a neural network. Hence, the input set and the output enclosure are "shadows" of the same latent space that we can use to transfer constraints. We present an efficient verification tool for neural networks that uses our iterative refinement to significantly reduce the number of subproblems in a branch-and-bound procedure. Using zonotopes as a set representation, unlike many other state-of-the-art approaches, our approach can be realized by only using matrix operations, which enables a significant speed-up through efficient GPU acceleration. We demonstrate that our tool achieves competitive performance, which would place it among the top-ranking tools of the last neural network verification competition (VNN-COMP'24). | [
{
"version": "v1",
"created": "Fri, 23 May 2025 13:05:07 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 12:14:26 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Koller",
"Lukas",
""
],
[
"Ladner",
"Tobias",
""
],
[
"Althoff",
"Matthias",
""
]
] |
2505.17871 | Zezhi Shao | Zezhi Shao, Yujie Li, Fei Wang, Chengqing Yu, Yisong Fu, Tangwen Qian, Bin Xu, Boyu Diao, Yongjun Xu, Xueqi Cheng | BLAST: Balanced Sampling Time Series Corpus for Universal Forecasting Models | Accepted by SIGKDD 2025 (Research Track) | null | 10.1145/3711896.3736860 | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | The advent of universal time series forecasting models has revolutionized zero-shot forecasting across diverse domains, yet the critical role of data diversity in training these models remains underexplored. Existing large-scale time series datasets often suffer from inherent biases and imbalanced distributions, leading to suboptimal model performance and generalization. To address this gap, we introduce BLAST, a novel pre-training corpus designed to enhance data diversity through a balanced sampling strategy. First, BLAST incorporates 321 billion observations from publicly available datasets and employs a comprehensive suite of statistical metrics to characterize time series patterns. Then, to facilitate pattern-oriented sampling, the data is implicitly clustered using grid-based partitioning. Furthermore, by integrating grid sampling and grid mixup techniques, BLAST ensures a balanced and representative coverage of diverse patterns. Experimental results demonstrate that models pre-trained on BLAST achieve state-of-the-art performance with a fraction of the computational resources and training tokens required by existing methods. Our findings highlight the pivotal role of data diversity in improving both training efficiency and model performance for the universal forecasting task. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 13:20:47 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 03:26:51 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Shao",
"Zezhi",
""
],
[
"Li",
"Yujie",
""
],
[
"Wang",
"Fei",
""
],
[
"Yu",
"Chengqing",
""
],
[
"Fu",
"Yisong",
""
],
[
"Qian",
"Tangwen",
""
],
[
"Xu",
"Bin",
""
],
[
"Diao",
"Boyu",
""
],
[
"Xu",
"Yongjun",
""
],
[
"Cheng",
"Xueqi",
""
]
] |
2505.17872 | Licheng Pan | Licheng Pan, Zhichao Chen, Haoxuan Li, Guangyi Liu, Zhijian Xu, Zhaoran Liu, Hao Wang, Ying Wei | Mixture of Low Rank Adaptation with Partial Parameter Sharing for Time Series Forecasting | null | null | null | null | cs.LG cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Multi-task forecasting has become the standard approach for time-series forecasting (TSF). However, we show that it suffers from an Expressiveness Bottleneck, where predictions at different time steps share the same representation, leading to unavoidable errors even with optimal representations. To address this issue, we propose a two-stage framework: first, pre-train a foundation model for one-step-ahead prediction; then, adapt it using step-specific LoRA modules.This design enables the foundation model to handle any number of forecast steps while avoiding the expressiveness bottleneck. We further introduce the Mixture-of-LoRA (MoLA) model, which employs adaptively weighted LoRA experts to achieve partial parameter sharing across steps. This approach enhances both efficiency and forecasting performance by exploiting interdependencies between forecast steps. Experiments show that MoLA significantly improves model expressiveness and outperforms state-of-the-art time-series forecasting methods. Code is available at https://anonymous.4open.science/r/MoLA-BC92. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 13:24:39 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 07:23:28 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Pan",
"Licheng",
""
],
[
"Chen",
"Zhichao",
""
],
[
"Li",
"Haoxuan",
""
],
[
"Liu",
"Guangyi",
""
],
[
"Xu",
"Zhijian",
""
],
[
"Liu",
"Zhaoran",
""
],
[
"Wang",
"Hao",
""
],
[
"Wei",
"Ying",
""
]
] |
2505.17939 | Manuel Lecha | Manuel Lecha, Andrea Cavallo, Francesca Dominici, Ran Levi, Alessio Del Bue, Elvin Isufi, Pietro Morerio, Claudio Battiloro | Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Graph Neural Networks (GNNs) excel at learning from pairwise interactions but often overlook multi-way and hierarchical relationships. Topological Deep Learning (TDL) addresses this limitation by leveraging combinatorial topological spaces. However, existing TDL models are restricted to undirected settings and fail to capture the higher-order directed patterns prevalent in many complex systems, e.g., brain networks, where such interactions are both abundant and functionally significant. To fill this gap, we introduce Semi-Simplicial Neural Networks (SSNs), a principled class of TDL models that operate on semi-simplicial sets -- combinatorial structures that encode directed higher-order motifs and their directional relationships. To enhance scalability, we propose Routing-SSNs, which dynamically select the most informative relations in a learnable manner. We prove that SSNs are strictly more expressive than standard graph and TDL models. We then introduce a new principled framework for brain dynamics representation learning, grounded in the ability of SSNs to provably recover topological descriptors shown to successfully characterize brain activity. Empirically, SSNs achieve state-of-the-art performance on brain dynamics classification tasks, outperforming the second-best model by up to 27%, and message passing GNNs by up to 50% in accuracy. Our results highlight the potential of principled topological models for learning from structured brain data, establishing a unique real-world case study for TDL. We also test SSNs on standard node classification and edge regression tasks, showing competitive performance. We will make the code and data publicly available. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 14:15:44 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 11:14:31 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Lecha",
"Manuel",
""
],
[
"Cavallo",
"Andrea",
""
],
[
"Dominici",
"Francesca",
""
],
[
"Levi",
"Ran",
""
],
[
"Del Bue",
"Alessio",
""
],
[
"Isufi",
"Elvin",
""
],
[
"Morerio",
"Pietro",
""
],
[
"Battiloro",
"Claudio",
""
]
] |
2505.17988 | YuTong Chen | Yutong Chen, Jiandong Gao, Ji Wu | Towards Revealing the Effectiveness of Small-Scale Fine-tuning in R1-style Reinforcement Learning | 11 figs, 3 table, preprint | null | null | null | cs.LG cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | R1-style Reinforcement Learning (RL) significantly enhances Large Language Models' reasoning capabilities, yet the mechanism behind rule-based RL remains unclear. We found that small-scale SFT has significant influence on RL but shows poor efficiency. To explain our observations, we propose an analytical framework and compare the efficiency of SFT and RL by measuring sample effect. Hypothetical analysis show that SFT efficiency is limited by training data. Guided by our analysis, we propose Re-distillation, a technique that fine-tunes pretrain model through small-scale distillation from the RL-trained policy. Experiments on Knight & Knave and MATH datasets demonstrate re-distillation's surprising efficiency: re-distilled models match RL performance with far fewer samples and less computation. Empirical verification shows that sample effect is a good indicator of performance improvements. As a result, on K&K dataset, our re-distilled Qwen2.5-1.5B model surpasses DeepSeek-V3-0324 with only 1K SFT samples. On MATH, Qwen2.5-1.5B fine-tuned with re-distilled 500 samples matches its instruct-tuned variant without RL. Our work explains several interesting phenomena in R1-style RL, shedding light on the mechanisms behind its empirical success. Code is available at: https://github.com/on1262/deep-reasoning | [
{
"version": "v1",
"created": "Fri, 23 May 2025 14:55:22 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:01:36 GMT"
}
] | 2025-05-28T00:00:00 | [
[
"Chen",
"Yutong",
""
],
[
"Gao",
"Jiandong",
""
],
[
"Wu",
"Ji",
""
]
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.