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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2505.17226 | Kun Yang | Kun Yang, Neena Imam | Secure and Private Federated Learning: Achieving Adversarial Resilience through Robust Aggregation | null | null | null | null | cs.LG cs.CR | http://creativecommons.org/licenses/by/4.0/ | Federated Learning (FL) enables collaborative machine learning across decentralized data sources without sharing raw data. It offers a promising approach to privacy-preserving AI. However, FL remains vulnerable to adversarial threats from malicious participants, referred to as Byzantine clients, who can send misleading updates to corrupt the global model. Traditional aggregation methods, such as simple averaging, are not robust to such attacks. More resilient approaches, like the Krum algorithm, require prior knowledge of the number of malicious clients, which is often unavailable in real-world scenarios. To address these limitations, we propose Average-rKrum (ArKrum), a novel aggregation strategy designed to enhance both the resilience and privacy guarantees of FL systems. Building on our previous work (rKrum), ArKrum introduces two key innovations. First, it includes a median-based filtering mechanism that removes extreme outliers before estimating the number of adversarial clients. Second, it applies a multi-update averaging scheme to improve stability and performance, particularly when client data distributions are not identical. We evaluate ArKrum on benchmark image and text datasets under three widely studied Byzantine attack types. Results show that ArKrum consistently achieves high accuracy and stability. It performs as well as or better than other robust aggregation methods. These findings demonstrate that ArKrum is an effective and practical solution for secure FL systems in adversarial environments. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 19:01:09 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 21:06:36 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Yang",
"Kun",
""
],
[
"Imam",
"Neena",
""
]
] |
2505.17427 | Zhengyi Zhao | Zhengyi Zhao, Shubo Zhang, Zezhong Wang, Huimin Wang, Yutian Zhao, Bin Liang, Yefeng Zheng, Binyang Li, Kam-Fai Wong, Xian Wu | T$^2$: An Adaptive Test-Time Scaling Strategy for Contextual Question Answering | arXiv admin note: substantial text overlap with arXiv:2503.22985 | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Recent advances in Large Language Models (LLMs) have demonstrated remarkable performance in Contextual Question Answering (CQA). However, prior approaches typically employ elaborate reasoning strategies regardless of question complexity, leading to low adaptability. Recent efficient test-time scaling methods introduce budget constraints or early stop mechanisms to avoid overthinking for straightforward questions. But they add human bias to the reasoning process and fail to leverage models' inherent reasoning capabilities. To address these limitations, we present T$^2$: Think-to-Think, a novel framework that dynamically adapts reasoning depth based on question complexity. T$^2$ leverages the insight that if an LLM can effectively solve similar questions using specific reasoning strategies, it can apply the same strategy to the original question. This insight enables to adoption of concise reasoning for straightforward questions while maintaining detailed analysis for complex problems. T$^2$ works through four key steps: decomposing questions into structural elements, generating similar examples with candidate reasoning strategies, evaluating these strategies against multiple criteria, and applying the most appropriate strategy to the original question. Experimental evaluation across seven diverse CQA benchmarks demonstrates that T$^2$ not only achieves higher accuracy than baseline methods but also reduces computational overhead by up to 25.2\%. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 03:18:02 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 08:58:33 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Zhao",
"Zhengyi",
""
],
[
"Zhang",
"Shubo",
""
],
[
"Wang",
"Zezhong",
""
],
[
"Wang",
"Huimin",
""
],
[
"Zhao",
"Yutian",
""
],
[
"Liang",
"Bin",
""
],
[
"Zheng",
"Yefeng",
""
],
[
"Li",
"Binyang",
""
],
[
"Wong",
"Kam-Fai",
""
],
[
"Wu",
"Xian",
""
]
] |
2505.17433 | Zhengyi Zhao | Zhengyi Zhao, Shubo Zhang, Yuxi Zhang, Yanxi Zhao, Yifan Zhang, Zezhong Wang, Huimin Wang, Yutian Zhao, Bin Liang, Yefeng Zheng, Binyang Li, Kam-Fai Wong, Xian Wu | MemeReaCon: Probing Contextual Meme Understanding in Large Vision-Language Models | null | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Memes have emerged as a popular form of multimodal online communication, where their interpretation heavily depends on the specific context in which they appear. Current approaches predominantly focus on isolated meme analysis, either for harmful content detection or standalone interpretation, overlooking a fundamental challenge: the same meme can express different intents depending on its conversational context. This oversight creates an evaluation gap: although humans intuitively recognize how context shapes meme interpretation, Large Vision Language Models (LVLMs) can hardly understand context-dependent meme intent. To address this critical limitation, we introduce MemeReaCon, a novel benchmark specifically designed to evaluate how LVLMs understand memes in their original context. We collected memes from five different Reddit communities, keeping each meme's image, the post text, and user comments together. We carefully labeled how the text and meme work together, what the poster intended, how the meme is structured, and how the community responded. Our tests with leading LVLMs show a clear weakness: models either fail to interpret critical information in the contexts, or overly focus on visual details while overlooking communicative purpose. MemeReaCon thus serves both as a diagnostic tool exposing current limitations and as a challenging benchmark to drive development toward more sophisticated LVLMs of the context-aware understanding. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 03:27:23 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 08:55:43 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Zhao",
"Zhengyi",
""
],
[
"Zhang",
"Shubo",
""
],
[
"Zhang",
"Yuxi",
""
],
[
"Zhao",
"Yanxi",
""
],
[
"Zhang",
"Yifan",
""
],
[
"Wang",
"Zezhong",
""
],
[
"Wang",
"Huimin",
""
],
[
"Zhao",
"Yutian",
""
],
[
"Liang",
"Bin",
""
],
[
"Zheng",
"Yefeng",
""
],
[
"Li",
"Binyang",
""
],
[
"Wong",
"Kam-Fai",
""
],
[
"Wu",
"Xian",
""
]
] |
2505.17836 | Anna Van Elst | Anna Van Elst, Igor Colin, Stephan Cl\'emen\c{c}on | Robust Distributed Estimation: Extending Gossip Algorithms to Ranking and Trimmed Means | null | null | null | null | stat.ML cs.LG stat.AP | http://creativecommons.org/licenses/by/4.0/ | This paper addresses the problem of robust estimation in gossip algorithms over arbitrary communication graphs. Gossip algorithms are fully decentralized, relying only on local neighbor-to-neighbor communication, making them well-suited for situations where communication is constrained. A fundamental challenge in existing mean-based gossip algorithms is their vulnerability to malicious or corrupted nodes. In this paper, we show that an outlier-robust mean can be computed by globally estimating a robust statistic. More specifically, we propose a novel gossip algorithm for rank estimation, referred to as \textsc{GoRank}, and leverage it to design a gossip procedure dedicated to trimmed mean estimation, coined \textsc{GoTrim}. In addition to a detailed description of the proposed methods, a key contribution of our work is a precise convergence analysis: we establish an $\mathcal{O}(1/t)$ rate for rank estimation and an $\mathcal{O}(\log(t)/\sqrt{t})$ rate for trimmed mean estimation, where by $t$ is meant the number of iterations. Moreover, we provide a breakdown point analysis of \textsc{GoTrim}. We empirically validate our theoretical results through experiments on diverse network topologies, data distributions and contamination schemes. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 12:51:03 GMT"
},
{
"version": "v2",
"created": "Fri, 30 May 2025 16:07:48 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 14:26:25 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Van Elst",
"Anna",
""
],
[
"Colin",
"Igor",
""
],
[
"Clémençon",
"Stephan",
""
]
] |
2505.18400 | Todd Brun | Juan Garcia Nila and Todd A. Brun | Continuous quantum correction on Markovian and Non-Markovian models | 21 pages, 15 figures; fixed grant number and a few minor typos | null | null | null | quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We investigate continuous quantum error correction, comparing performance under a Markovian error model to two distinct non-Markovian models. The first non-Markovian model involves an interaction Hamiltonian between the system and an environmental qubit via an X-X coupling, with a "cooling" bath acting on the environment qubit. This model is known to exhibit abrupt transitions between Markovian and non-Markovian behavior. The second non-Markovian model uses the post-Markovian master equation (PMME), which represents the bath correlation through a memory kernel; we consider an exponentially decaying kernel and both underdamped and overdamped dynamics. We systematically compare these non-Markovian error models against the Markovian case and against each other, for a variety of different codes. We start with a single qubit, which can be solved analytically. We then consider the three-qubit repetition code and the five-qubit "perfect" code. In all cases, we find that the fidelity decays more rapidly in the Markovian case than in either non-Markovian model, suggesting that continuous quantum error correction has enhanced performance against non-Markovian noise. We attribute this difference to the presence of a quantum Zeno regime in both non-Markovian models. | [
{
"version": "v1",
"created": "Fri, 23 May 2025 22:08:32 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 21:40:40 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Nila",
"Juan Garcia",
""
],
[
"Brun",
"Todd A.",
""
]
] |
2505.18570 | Mohammad Erfan Sadeghi | Tina Khezresmaeilzadeh, Parsa Razmara, Seyedarmin Azizi, Mohammad Erfan Sadeghi, Erfan Baghaei Potraghloo | VISTA: Vision-Language Inference for Training-Free Stock Time-Series Analysis | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Stock price prediction remains a complex and high-stakes task in financial analysis, traditionally addressed using statistical models or, more recently, language models. In this work, we introduce VISTA (Vision-Language Inference for Stock Time-series Analysis), a novel, training-free framework that leverages Vision-Language Models (VLMs) for multi-modal stock forecasting. VISTA prompts a VLM with both textual representations of historical stock prices and their corresponding line charts to predict future price values. By combining numerical and visual modalities in a zero-shot setting and using carefully designed chain-of-thought prompts, VISTA captures complementary patterns that unimodal approaches often miss. We benchmark VISTA against standard baselines, including ARIMA and text-only LLM-based prompting methods. Experimental results show that VISTA outperforms these baselines by up to 89.83%, demonstrating the effectiveness of multi-modal inference for stock time-series analysis and highlighting the potential of VLMs in financial forecasting tasks without requiring task-specific training. | [
{
"version": "v1",
"created": "Sat, 24 May 2025 07:20:14 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 00:30:58 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Khezresmaeilzadeh",
"Tina",
""
],
[
"Razmara",
"Parsa",
""
],
[
"Azizi",
"Seyedarmin",
""
],
[
"Sadeghi",
"Mohammad Erfan",
""
],
[
"Potraghloo",
"Erfan Baghaei",
""
]
] |
2505.18776 | Gleb Gribakin | R. S. Wilde, G. F. Gribakin, and I. I. Fabrikant | Dissociative positronium attachment in halogen gases | 25 pages, 7 figures, accepted for publication in Phys. Rev. A | Phys. Rev. A 111, 052817 (2025) | 10.1103/PhysRevA.111.052817 | null | physics.atom-ph | http://creativecommons.org/licenses/by/4.0/ | We suggest that the observed large annihilation rates of ortho-positronium ($o$-Ps) in halogen gases are due to the process of dissociative Ps attachment, ${\rm Ps} + X_2 \to {\rm Ps}X + X$, where $X$ stands for a halogen atom. This process is similar to dissociative electron attachment which leads to formation of negative ions. We calculate the cross section and rate of this process for the F$_2$ molecule, for which it is exothermic, and therefore, can occur at room temperature. We start with the Ps-F$_2$ scattering calculations which take into account electron exchange and correlations within the framework of the free-electron-gas model. The calculations reveal several resonances. Similar to the process of dissociative electron attachment, a $\Sigma_u$ resonance contributes to the dissociative Ps attachment at thermal energies. We determine the resonance position and width as functions of the internuclear separation, and use them as inputs for the local version of the quasiclassical theory of dissociative attachment. Our calculations yield an anomalously large rate constant for the $o$-Ps annihilation process which is only one order of magnitude lower than those observed for Br$_2$ and I$_2$. | [
{
"version": "v1",
"created": "Sat, 24 May 2025 16:29:38 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Wilde",
"R. S.",
""
],
[
"Gribakin",
"G. F.",
""
],
[
"Fabrikant",
"I. I.",
""
]
] |
2505.19052 | Seokcheon Lee | Seokcheon Lee | Probing Time-Varying Dark Energy with DESI: The Crucial Role of Precision Matter Density ($\Omega_{m0}$) Measurements | 33 pages, 6 figures, update references | null | null | null | astro-ph.CO | http://creativecommons.org/licenses/by/4.0/ | Accurate measurements of fundamental cosmological parameters, especially the Hubble constant ($H_0$) and present-day matter density ($\Omega_{m0}$), are crucial for constraining dark energy (DE) evolution. We analyze the sensitivities of cosmological observables ($H(z)$, $D_L(z)$, $\EG$) to $\Omega_{m0}$, $\omega_0$, and $\omega_a$ under different parametrizations. Our results show observables are far more sensitive to $\Omega_{m0}$ than to DE equation of state parameters (e.g., at $z \sim 0.5$, $H(z)$'s $\Omega_{m0}$ sensitivity is $\sim 0.7$ vs. $\omega_a$'s $\sim 0.04$). This hierarchy mandates high-precision $\Omega_{m0}$ measurements to accurately constrain time-varying DE. We also find DE parameter sensitivity highly depends on parametrization; the standard CPL form shows low sensitivity to $\omega_a$, but $\omega(z) = \omega_0 + \omega_a \ln(1+z)$ significantly enhances it. Our analysis of DESI DR1/DR2 data confirms these theoretical limits: standalone DESI data primarily provides only upper limits for $\omega_a$, underscoring insufficient constraining power for a definitive time-varying DE detection. While combined datasets offer tighter constraints, interpretation requires caution due to parametrization influence. In conclusion, improving $\Omega_{m0}$ precision and adopting optimized parametrizations are imperative for future surveys like DESI to fully probe dark energy's nature. | [
{
"version": "v1",
"created": "Sun, 25 May 2025 09:13:58 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 14:34:54 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Lee",
"Seokcheon",
""
]
] |
2505.19176 | Zhuo Liu | Zhuo Liu, Moxin Li, Xun Deng, Qifan Wang, Fuli Feng | Assistant-Guided Mitigation of Teacher Preference Bias in LLM-as-a-Judge | Under review | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | LLM-as-a-Judge employs large language models (LLMs), such as GPT-4, to evaluate the quality of LLM-generated responses, gaining popularity for its cost-effectiveness and strong alignment with human evaluations. However, training proxy judge models using evaluation data generated by powerful teacher models introduces a critical yet previously overlooked issue: teacher preference bias, where the proxy judge model learns a biased preference for responses from the teacher model. To tackle this problem, we propose a novel setting that incorporates an additional assistant model, which is not biased toward the teacher model's responses, to complement the training data. Building on this setup, we introduce AGDe-Judge, a three-stage framework designed to debias from both the labels and feedbacks in the training data. Extensive experiments demonstrate that AGDe-Judge effectively reduces teacher preference bias while maintaining strong performance across six evaluation benchmarks. Code is available at https://github.com/Liuz233/AGDe-Judge. | [
{
"version": "v1",
"created": "Sun, 25 May 2025 14:48:49 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 16:16:31 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Liu",
"Zhuo",
""
],
[
"Li",
"Moxin",
""
],
[
"Deng",
"Xun",
""
],
[
"Wang",
"Qifan",
""
],
[
"Feng",
"Fuli",
""
]
] |
2505.19412 | Nagare Katayama | Nagare Katayama, Yuya Tanizaki | 2d Cardy-Rabinovici model with the modified Villain lattice: Exact dualities and symmetries | 28 pages, 4 figures | null | null | YITP-25-78 | hep-th hep-lat | http://creativecommons.org/licenses/by/4.0/ | The Cardy-Rabinovici model is a toy model of the lattice $U(1)$ gauge theories to study various oblique confinement states associated with the nonzero $\theta$ angles. We reformulate the $2$d version of this model using the modified Villain lattice formalism, and we establish the exact $\theta$ periodicity for the Witten effect and the strong-weak duality at the finite lattice spacings. We then study the phase structure of this model based on the duality, symmetry and anomaly, and the perturbative renormalization group. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 02:04:31 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 09:04:46 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Katayama",
"Nagare",
""
],
[
"Tanizaki",
"Yuya",
""
]
] |
2505.19449 | Andrei Dubikovsky | A.I. Dubikovsky, P.K. Silaev | Simple finite-dimensional model of the metastable state | null | Moscow University Physics Bulletin, 2025, Vol. 80, No. 2, pp. 226-235 | 10.3103/S0027134925700298 | null | quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We have constructed an approximate analytical solution of the spectral problem for a finite-dimensional matrix of a special kind, which turns out to be a very simple and quite satisfactory model of the metastable state. Most of the characteristic properties of the metastable state are reproduced: line shape, decay dynamics, and density of states. The correctness of the approximate analytical solution was verified by direct numerical calculations. The proposed model is a finite-dimensional analog of the Fano formalism. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 03:15:52 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Dubikovsky",
"A. I.",
""
],
[
"Silaev",
"P. K.",
""
]
] |
2505.19641 | Junteng Liu | Junteng Liu, Yuanxiang Fan, Zhuo Jiang, Han Ding, Yongyi Hu, Chi Zhang, Yiqi Shi, Shitong Weng, Aili Chen, Shiqi Chen, Yunan Huang, Mozhi Zhang, Pengyu Zhao, Junjie Yan, Junxian He | SynLogic: Synthesizing Verifiable Reasoning Data at Scale for Learning Logical Reasoning and Beyond | null | null | null | null | cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Recent advances such as OpenAI-o1 and DeepSeek R1 have demonstrated the potential of Reinforcement Learning (RL) to enhance reasoning abilities in Large Language Models (LLMs). While open-source replication efforts have primarily focused on mathematical and coding domains, methods and resources for developing general reasoning capabilities remain underexplored. This gap is partly due to the challenge of collecting diverse and verifiable reasoning data suitable for RL. We hypothesize that logical reasoning is critical for developing general reasoning capabilities, as logic forms a fundamental building block of reasoning. In this work, we present SynLogic, a data synthesis framework and dataset that generates diverse logical reasoning data at scale, encompassing 35 diverse logical reasoning tasks. The SynLogic approach enables controlled synthesis of data with adjustable difficulty and quantity. Importantly, all examples can be verified by simple rules, making them ideally suited for RL with verifiable rewards. In our experiments, we validate the effectiveness of RL training on the SynLogic dataset based on 7B and 32B models. SynLogic leads to state-of-the-art logical reasoning performance among open-source datasets, surpassing DeepSeek-R1-Distill-Qwen-32B by 6 points on BBEH. Furthermore, mixing SynLogic data with mathematical and coding tasks improves the training efficiency of these domains and significantly enhances reasoning generalization. Notably, our mixed training model outperforms DeepSeek-R1-Zero-Qwen-32B across multiple benchmarks. These findings position SynLogic as a valuable resource for advancing the broader reasoning capabilities of LLMs. We open-source both the data synthesis pipeline and the SynLogic dataset at https://github.com/MiniMax-AI/SynLogic. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 07:59:36 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 03:03:09 GMT"
},
{
"version": "v3",
"created": "Wed, 28 May 2025 16:04:03 GMT"
},
{
"version": "v4",
"created": "Wed, 04 Jun 2025 05:08:08 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Liu",
"Junteng",
""
],
[
"Fan",
"Yuanxiang",
""
],
[
"Jiang",
"Zhuo",
""
],
[
"Ding",
"Han",
""
],
[
"Hu",
"Yongyi",
""
],
[
"Zhang",
"Chi",
""
],
[
"Shi",
"Yiqi",
""
],
[
"Weng",
"Shitong",
""
],
[
"Chen",
"Aili",
""
],
[
"Chen",
"Shiqi",
""
],
[
"Huang",
"Yunan",
""
],
[
"Zhang",
"Mozhi",
""
],
[
"Zhao",
"Pengyu",
""
],
[
"Yan",
"Junjie",
""
],
[
"He",
"Junxian",
""
]
] |
2505.19653 | Liu Yibo | Ning Yang, Hai Lin, Yibo Liu, Baoliang Tian, Guoqing Liu, Haijun Zhang | Token-Importance Guided Direct Preference Optimization | null | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Ensuring that large language models (LLMs) generate outputs aligned with human preferences is important for safe and effective AI interactions. While Direct Preference Optimization (DPO) employs an implicit reward function to optimize the policy model, however, it and its related variants overlook the differential importance of individual tokens and are sensitive to judgment noise in preference datasets during generation. Although recent methods attempt to assess the important weight of tokens via probability prediction or simplistic weighting schemes, these evaluation methods are prone to biases and still cannot fully address these issues. To solve this problem, we propose the Token-Importance Guided Direct Preference Optimization (TI-DPO), which introduces two key innovations: the gradient-based token-importance weights that dynamically prioritize critical tokens, and a triple loss that explicitly guides model outputs to approach human-preferred responses and stay away from non-preferred responses. Experimental results show that TI-DPO achieves higher accuracy and stronger generative diversity, providing more stable and computationally efficient solutions compared with DPO and other RLHF methods. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 08:11:24 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Yang",
"Ning",
""
],
[
"Lin",
"Hai",
""
],
[
"Liu",
"Yibo",
""
],
[
"Tian",
"Baoliang",
""
],
[
"Liu",
"Guoqing",
""
],
[
"Zhang",
"Haijun",
""
]
] |
2505.19931 | Qixi Zheng | Qixi Zheng, Yushen Chen, Zhikang Niu, Ziyang Ma, Xiaofei Wang, Kai Yu, Xie Chen | Accelerating Flow-Matching-Based Text-to-Speech via Empirically Pruned Step Sampling | null | null | null | null | eess.AS cs.SD | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Flow-matching-based text-to-speech (TTS) models, such as Voicebox, E2 TTS, and F5-TTS, have attracted significant attention in recent years. These models require multiple sampling steps to reconstruct speech from noise, making inference speed a key challenge. Reducing the number of sampling steps can greatly improve inference efficiency. To this end, we introduce Fast F5-TTS, a training-free approach to accelerate the inference of flow-matching-based TTS models. By inspecting the sampling trajectory of F5-TTS, we identify redundant steps and propose Empirically Pruned Step Sampling (EPSS), a non-uniform time-step sampling strategy that effectively reduces the number of sampling steps. Our approach achieves a 7-step generation with an inference RTF of 0.030 on an NVIDIA RTX 3090 GPU, making it 4 times faster than the original F5-TTS while maintaining comparable performance. Furthermore, EPSS performs well on E2 TTS models, demonstrating its strong generalization ability. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 12:58:27 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 11:58:27 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Zheng",
"Qixi",
""
],
[
"Chen",
"Yushen",
""
],
[
"Niu",
"Zhikang",
""
],
[
"Ma",
"Ziyang",
""
],
[
"Wang",
"Xiaofei",
""
],
[
"Yu",
"Kai",
""
],
[
"Chen",
"Xie",
""
]
] |
2505.19945 | Gangshan Jing | Jinpeng Huang, Gangshan Jing | Signed Angle Rigid Graphs for Network Localization and Formation Control | null | null | null | null | math.OC cs.SY eess.SY | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Graph rigidity theory studies the capability of a graph embedded in the Euclidean space to constrain its global geometric shape via local constraints among nodes and edges, and has been widely exploited in network localization and formation control. In recent years, the traditional rigidity theory has been extended by considering new types of local constraints such as bearing, angle, ratio of distance, etc. Among them, the signed angle constraint has received extensive attention, since it is practically measurable and independent of the global coordinate frame. However, the relevant studies always consider special graph structures, which are sufficient but not necessary for signed angle rigidity. This paper presents a comprehensive combinatorial analysis in terms of graphs and angle index sets for signed angle rigidity. We show that Laman graphs equivalently characterize minimally signed angle rigid graphs. Moreover, we propose a method to construct the minimal set of signed angle constraints in a Laman graph to effectively ensure signed angle rigidity. These results are finally applied to distributed network localization and formation stabilization problems, respectively, where each agent only has access to signed angle measurements. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 13:10:08 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 07:50:31 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Huang",
"Jinpeng",
""
],
[
"Jing",
"Gangshan",
""
]
] |
2505.20015 | Ramon Ferrer-i-Cancho | Ramon Ferrer-i-Cancho | On the class of coding optimality of human languages and the origins of Zipf's law | null | null | null | null | cs.CL physics.soc-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Here we present a new class of optimality for coding systems. Members of that class are displaced linearly from optimal coding and thus exhibit Zipf's law, namely a power-law distribution of frequency ranks. Within that class, Zipf's law, the size-rank law and the size-probability law form a group-like structure. We identify human languages that are members of the class. All languages showing sufficient agreement with Zipf's law are potential members of the class. In contrast, there are communication systems in other species that cannot be members of that class for exhibiting an exponential distribution instead but dolphins and humpback whales might. We provide a new insight into plots of frequency versus rank in double logarithmic scale. For any system, a straight line in that scale indicates that the lengths of optimal codes under non-singular coding and under uniquely decodable encoding are displaced by a linear function whose slope is the exponent of Zipf's law. For systems under compression and constrained to be uniquely decodable, such a straight line may indicate that the system is coding close to optimality. We provide support for the hypothesis that Zipf's law originates from compression and define testable conditions for the emergence of Zipf's law in compressing systems. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 14:05:45 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 17:00:20 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 11:35:43 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Ferrer-i-Cancho",
"Ramon",
""
]
] |
2505.20040 | Ryota Maeda | Yasutaka Koga, Ryota Maeda, Daiki Saito, Keiya Uemichi, and Daisuke Yoshida | Dynamical Formation of Charged Wormholes | 34 pages, 10 figures | null | null | null | gr-qc hep-th | http://creativecommons.org/licenses/by/4.0/ | We construct static, spherically symmetric, charged traversable wormhole solutions to the Einstein--Maxwell equations, supported by bidirectional (ingoing and outgoing) null dust with negative energy, and discuss a scenario for their dynamical formation from a black hole. Our solution contains a traversable throat, where the areal radius takes a minimum, although the spacetime is not asymptotically flat. In our formation scenario, the spacetime evolves sequentially from a black hole to Vaidya regions and finally to a wormhole, with each transition mediated by an impulsive null shell. We find that the radius of the wormhole throat is determined by the mass and charge of the initial black hole as well as those of the injected shell. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 14:25:09 GMT"
},
{
"version": "v2",
"created": "Wed, 28 May 2025 06:01:28 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Koga",
"Yasutaka",
""
],
[
"Maeda",
"Ryota",
""
],
[
"Saito",
"Daiki",
""
],
[
"Uemichi",
"Keiya",
""
],
[
"Yoshida",
"Daisuke",
""
]
] |
2505.20290 | Vincent Liu | Vincent Liu, Ademi Adeniji, Haotian Zhan, Siddhant Haldar, Raunaq Bhirangi, Pieter Abbeel, Lerrel Pinto | EgoZero: Robot Learning from Smart Glasses | null | null | null | null | cs.RO cs.AI | http://creativecommons.org/licenses/by/4.0/ | Despite recent progress in general purpose robotics, robot policies still lag far behind basic human capabilities in the real world. Humans interact constantly with the physical world, yet this rich data resource remains largely untapped in robot learning. We propose EgoZero, a minimal system that learns robust manipulation policies from human demonstrations captured with Project Aria smart glasses, $\textbf{and zero robot data}$. EgoZero enables: (1) extraction of complete, robot-executable actions from in-the-wild, egocentric, human demonstrations, (2) compression of human visual observations into morphology-agnostic state representations, and (3) closed-loop policy learning that generalizes morphologically, spatially, and semantically. We deploy EgoZero policies on a gripper Franka Panda robot and demonstrate zero-shot transfer with 70% success rate over 7 manipulation tasks and only 20 minutes of data collection per task. Our results suggest that in-the-wild human data can serve as a scalable foundation for real-world robot learning - paving the way toward a future of abundant, diverse, and naturalistic training data for robots. Code and videos are available at https://egozero-robot.github.io. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 17:59:17 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 22:50:28 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Liu",
"Vincent",
""
],
[
"Adeniji",
"Ademi",
""
],
[
"Zhan",
"Haotian",
""
],
[
"Haldar",
"Siddhant",
""
],
[
"Bhirangi",
"Raunaq",
""
],
[
"Abbeel",
"Pieter",
""
],
[
"Pinto",
"Lerrel",
""
]
] |
2505.20436 | Yigal Shamir | Maarten Golterman and Yigal Shamir | Constraints on the symmetric mass generation paradigm for lattice chiral gauge theories | RevTex, 37 pages | null | null | null | hep-lat cond-mat.str-el hep-ph hep-th | http://creativecommons.org/licenses/by/4.0/ | Within the symmetric mass generation (SMG) approach to the construction of lattice chiral gauge theories, one attempts to use interactions to render mirror fermions massive without symmetry breaking, thus obtaining the desired chiral massless spectrum. If successful, the gauge field can be turned on, and thus a chiral gauge theory can be constructed in the phase in which SMG takes place. In this paper we argue that the zeros that often replace the mirror poles of fermion two-point functions in an SMG phase are ``kinematical'' singularities, which can be avoided by choosing an appropriate set of interpolating fields that contains both elementary and composite fields. This allows us to apply general constraints on the existence of a chiral fermion spectrum which are valid in the presence of (non-gauge) interactions of arbitrary strength, including in any SMG phase. Using a suitably constructed one-particle lattice hamiltonian describing the fermion spectrum, we discuss the conditions for the applicability of the Nielsen-Ninomiya theorem to this hamiltonian. If these conditions are satisfied, the massless fermion spectrum must be vector-like. We add some general observations on the strong coupling limit of SMG models. Finally, we elaborate on the qualitative differences between four-dimensional and two-dimensional theories that limit the lessons that can be drawn from two-dimensional models. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 18:32:49 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 06:43:12 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Golterman",
"Maarten",
""
],
[
"Shamir",
"Yigal",
""
]
] |
2505.20459 | Dimitrios Giataganas | Dimitrios Giataganas | Holographic Timelike c-function | v2: 9 pages, double-column format; minor improvements | null | null | null | hep-th cond-mat.stat-mech | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The integration of high-energy degrees of freedom along the renormalization group (RG) flow in Poincar\'e-invariant theories can be captured by a monotonic c-function. For such theories, holographic monotonic c-functions have been constructed using entanglement entropy. However, in theories with broken Lorentz invariance, such constructions generally fail, reflecting both the violation of the entanglement RG monotonicity and its limitations in capturing certain properties of non-relativistic RG flows. Since many quantum many-body systems lack Lorentz invariance, it is of significant importance to identify a quantity that reflects the decrease in degrees of freedom along non-relativistic RG flows. We show that the recently introduced holographic timelike entanglement entropy naturally gives rise to a new c-function applicable to all such theories. We further demonstrate the existence of this c-function in theories with Lifshitz and hyperscaling-violating fixed points, showing that, provided the null energy conditions and thermodynamic stability are satisfied, the proposed c-function exhibits the expected monotonic behavior along the RG flow. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 18:54:40 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 13:53:40 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Giataganas",
"Dimitrios",
""
]
] |
2505.20538 | Sebastian Joseph | Sebastian Antony Joseph, Syed Murtaza Husain, Stella S. R. Offner, St\'ephanie Juneau, Paul Torrey, Adam S. Bolton, Juan P. Farias, Niall Gaffney, Greg Durrett, Junyi Jessy Li | AstroVisBench: A Code Benchmark for Scientific Computing and Visualization in Astronomy | null | null | null | null | cs.CL astro-ph.IM cs.LG | http://creativecommons.org/licenses/by/4.0/ | Large Language Models (LLMs) are being explored for applications in scientific research, including their capabilities to synthesize literature, answer research questions, generate research ideas, and even conduct computational experiments. Ultimately, our goal is for these to help scientists derive novel scientific insights. In many areas of science, such insights often arise from processing and visualizing data to understand its patterns. However, evaluating whether an LLM-mediated scientific workflow produces outputs conveying the correct scientific insights is challenging to evaluate and has not been addressed in past work. We introduce AstroVisBench, the first benchmark for both scientific computing and visualization in the astronomy domain. AstroVisBench judges a language model's ability to both (1) create astronomy-specific workflows to process and analyze data and (2) visualize the results of these workflows through complex plots. Our evaluation of visualizations uses a novel LLM-as-a-judge workflow, which is validated against annotation by five professional astronomers. Using AstroVisBench we present an evaluation of state-of-the-art language models, showing a significant gap in their ability to engage in astronomy research as useful assistants. This evaluation provides a strong end-to-end evaluation for AI scientists that offers a path forward for the development of visualization-based workflows, which are central to a broad range of domains from physics to biology. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 21:49:18 GMT"
},
{
"version": "v2",
"created": "Wed, 28 May 2025 14:54:54 GMT"
},
{
"version": "v3",
"created": "Tue, 03 Jun 2025 18:56:38 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Joseph",
"Sebastian Antony",
""
],
[
"Husain",
"Syed Murtaza",
""
],
[
"Offner",
"Stella S. R.",
""
],
[
"Juneau",
"Stéphanie",
""
],
[
"Torrey",
"Paul",
""
],
[
"Bolton",
"Adam S.",
""
],
[
"Farias",
"Juan P.",
""
],
[
"Gaffney",
"Niall",
""
],
[
"Durrett",
"Greg",
""
],
[
"Li",
"Junyi Jessy",
""
]
] |
2505.20573 | Jiabao Ji | Jiabao Ji, Yongchao Chen, Yang Zhang, Ramana Rao Kompella, Chuchu Fan, Gaowen Liu, Shiyu Chang | Collision- and Reachability-Aware Multi-Robot Control with Grounded LLM Planners | null | null | null | null | cs.RO cs.AI | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) have demonstrated strong performance in various robot control tasks. However, their deployment in real-world applications remains constrained. Even state-ofthe-art LLMs, such as GPT-o4mini, frequently produce invalid action plans that violate physical constraints, such as directing a robot to an unreachable location or causing collisions between robots. This issue primarily arises from a lack of awareness of these physical constraints during the reasoning process. To address this issue, we propose a novel framework that integrates reinforcement learning with verifiable rewards (RLVR) to incentivize knowledge of physical constraints into LLMs to induce constraints-aware reasoning during plan generation. In this approach, only valid action plans that successfully complete a control task receive positive rewards. We applied our method to two small-scale LLMs: a non-reasoning Qwen2.5-3B-Instruct and a reasoning Qwen3-4B. The experiment results demonstrate that constraint-aware small LLMs largely outperform large-scale models without constraints, grounded on both the BoxNet task and a newly developed BoxNet3D environment built using MuJoCo. This work highlights the effectiveness of grounding even small LLMs with physical constraints to enable scalable and efficient multi-robot control in complex, physically constrained environments. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 23:14:16 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 19:33:56 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Ji",
"Jiabao",
""
],
[
"Chen",
"Yongchao",
""
],
[
"Zhang",
"Yang",
""
],
[
"Kompella",
"Ramana Rao",
""
],
[
"Fan",
"Chuchu",
""
],
[
"Liu",
"Gaowen",
""
],
[
"Chang",
"Shiyu",
""
]
] |
2505.20645 | Zihao He | Kai Chen, Zihao He, Taiwei Shi, Kristina Lerman | STEER-BENCH: A Benchmark for Evaluating the Steerability of Large Language Models | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Steerability, or the ability of large language models (LLMs) to adapt outputs to align with diverse community-specific norms, perspectives, and communication styles, is critical for real-world applications but remains under-evaluated. We introduce Steer-Bench, a benchmark for assessing population-specific steering using contrasting Reddit communities. Covering 30 contrasting subreddit pairs across 19 domains, Steer-Bench includes over 10,000 instruction-response pairs and validated 5,500 multiple-choice question with corresponding silver labels to test alignment with diverse community norms. Our evaluation of 13 popular LLMs using Steer-Bench reveals that while human experts achieve an accuracy of 81% with silver labels, the best-performing models reach only around 65% accuracy depending on the domain and configuration. Some models lag behind human-level alignment by over 15 percentage points, highlighting significant gaps in community-sensitive steerability. Steer-Bench is a benchmark to systematically assess how effectively LLMs understand community-specific instructions, their resilience to adversarial steering attempts, and their ability to accurately represent diverse cultural and ideological perspectives. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 02:47:56 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 06:18:10 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Chen",
"Kai",
""
],
[
"He",
"Zihao",
""
],
[
"Shi",
"Taiwei",
""
],
[
"Lerman",
"Kristina",
""
]
] |
2505.20704 | Zixuan Hu | Zixuan Hu, Yichun Hu, Xiaotong Li, Shixiang Tang, Ling-Yu Duan | Beyond Entropy: Region Confidence Proxy for Wild Test-Time Adaptation | Accepted by ICML 2025 | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Wild Test-Time Adaptation (WTTA) is proposed to adapt a source model to unseen domains under extreme data scarcity and multiple shifts. Previous approaches mainly focused on sample selection strategies, while overlooking the fundamental problem on underlying optimization. Initially, we critically analyze the widely-adopted entropy minimization framework in WTTA and uncover its significant limitations in noisy optimization dynamics that substantially hinder adaptation efficiency. Through our analysis, we identify region confidence as a superior alternative to traditional entropy, however, its direct optimization remains computationally prohibitive for real-time applications. In this paper, we introduce a novel region-integrated method ReCAP that bypasses the lengthy process. Specifically, we propose a probabilistic region modeling scheme that flexibly captures semantic changes in embedding space. Subsequently, we develop a finite-to-infinite asymptotic approximation that transforms the intractable region confidence into a tractable and upper-bounded proxy. These innovations significantly unlock the overlooked potential dynamics in local region in a concise solution. Our extensive experiments demonstrate the consistent superiority of ReCAP over existing methods across various datasets and wild scenarios. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 04:30:13 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 07:00:00 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Hu",
"Zixuan",
""
],
[
"Hu",
"Yichun",
""
],
[
"Li",
"Xiaotong",
""
],
[
"Tang",
"Shixiang",
""
],
[
"Duan",
"Ling-Yu",
""
]
] |
2505.20730 | Shahrooz Pouryousef | Shahrooz Pouryousef, Ali Montazeralghaem | What LLMs Miss in Recommendations: Bridging the Gap with Retrieval-Augmented Collaborative Signals | null | null | null | null | cs.IR cs.AI cs.CL cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | User-item interactions contain rich collaborative signals that form the backbone of many successful recommender systems. While recent work has explored the use of large language models (LLMs) for recommendation, it remains unclear whether LLMs can effectively reason over this type of collaborative information. In this paper, we conduct a systematic comparison between LLMs and classical matrix factorization (MF) models to assess LLMs' ability to leverage user-item interaction data. We further introduce a simple retrieval-augmented generation (RAG) method that enhances LLMs by grounding their predictions in structured interaction data. Our experiments reveal that current LLMs often fall short in capturing collaborative patterns inherent to MF models, but that our RAG-based approach substantially improves recommendation quality-highlighting a promising direction for future LLM-based recommenders. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 05:18:57 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 00:54:43 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Pouryousef",
"Shahrooz",
""
],
[
"Montazeralghaem",
"Ali",
""
]
] |
2505.20734 | Cheng Zhuoyu | Zhuoyu Cheng, Kohei Hatano, Eiji Takimoto | Adversarial bandit optimization for approximately linear functions | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | We consider a bandit optimization problem for nonconvex and non-smooth functions, where in each trial the loss function is the sum of a linear function and a small but arbitrary perturbation chosen after observing the player's choice. We give both expected and high probability regret bounds for the problem. Our result also implies an improved high-probability regret bound for the bandit linear optimization, a special case with no perturbation. We also give a lower bound on the expected regret. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 05:22:01 GMT"
},
{
"version": "v2",
"created": "Sun, 01 Jun 2025 06:19:26 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 05:08:36 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Cheng",
"Zhuoyu",
""
],
[
"Hatano",
"Kohei",
""
],
[
"Takimoto",
"Eiji",
""
]
] |
2505.20875 | Jiyoung Lee | Jiyoung Lee, Seungho Kim, Jieun Han, Jun-Min Lee, Kitaek Kim, Alice Oh, Edward Choi | Trans-EnV: A Framework for Evaluating the Linguistic Robustness of LLMs Against English Varieties | 27 pages, 6 figures, 16 tables | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by-sa/4.0/ | Large Language Models (LLMs) are predominantly evaluated on Standard American English (SAE), often overlooking the diversity of global English varieties. This narrow focus may raise fairness concerns as degraded performance on non-standard varieties can lead to unequal benefits for users worldwide. Therefore, it is critical to extensively evaluate the linguistic robustness of LLMs on multiple non-standard English varieties. We introduce Trans-EnV, a framework that automatically transforms SAE datasets into multiple English varieties to evaluate the linguistic robustness. Our framework combines (1) linguistics expert knowledge to curate variety-specific features and transformation guidelines from linguistic literature and corpora, and (2) LLM-based transformations to ensure both linguistic validity and scalability. Using Trans-EnV, we transform six benchmark datasets into 38 English varieties and evaluate seven state-of-the-art LLMs. Our results reveal significant performance disparities, with accuracy decreasing by up to 46.3% on non-standard varieties. These findings highlight the importance of comprehensive linguistic robustness evaluation across diverse English varieties. Each construction of Trans-EnV was validated through rigorous statistical testing and consultation with a researcher in the field of second language acquisition, ensuring its linguistic validity. Our code and datasets are publicly available at https://github.com/jiyounglee-0523/TransEnV and https://huggingface.co/collections/jiyounglee0523/transenv-681eadb3c0c8cf363b363fb1. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 08:23:27 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 04:46:38 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Lee",
"Jiyoung",
""
],
[
"Kim",
"Seungho",
""
],
[
"Han",
"Jieun",
""
],
[
"Lee",
"Jun-Min",
""
],
[
"Kim",
"Kitaek",
""
],
[
"Oh",
"Alice",
""
],
[
"Choi",
"Edward",
""
]
] |
2505.20944 | Myroslava Hladchenko | Myroslava Hladchenko | International collaboration of Ukrainian scholars: Effects of Russia's full-scale invasion of Ukraine | null | null | null | null | cs.DL | http://creativecommons.org/licenses/by-nc-sa/4.0/ | This study explores the effects of Russia's full-scale invasion of Ukraine on the international collaboration of Ukrainian scholars. First and foremost, Ukrainian scholars deserve respect for continuing to publish despite life-threatening conditions, mental strain, shelling and blackouts. In 2022-2023, universities gained more from international collaboration than the NASU. The percentage of internationally co-authored articles remained unchanged for the NASU, while it increased for universities. In 2023, 40.8% of articles published by the NASU and 32,2% of articles published by universities were internationally co-authored. However, these figures are still much lower than in developed countries (60-70%). The citation impact of internationally co-authored articles remained statistically unchanged for the NASU but increased for universities. The highest share of internationally co-authored articles published by the NASU in both periods was in the physical sciences and engineering. However, the citation impact of these articles declined in 2022-2023, nearly erasing their previous citation advantage over university publications. Universities consistently outperformed the NASU in the citation impact of internationally co-authored articles in biomedical and health sciences across both periods. International collaboration can help Ukrainian scholars to go through this difficult time. In turn, they can contribute to the strengthening of Europe. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 09:28:21 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 14:55:52 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Hladchenko",
"Myroslava",
""
]
] |
2505.21082 | Jieyong Kim | Jieyong Kim, Tongyoung Kim, Soojin Yoon, Jaehyung Kim, Dongha Lee | LLMs Think, But Not In Your Flow: Reasoning-Level Personalization for Black-Box Large Language Models | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Large language models (LLMs) have recently achieved impressive performance across a wide range of natural language tasks and are now widely used in real-world applications. Among them, black-box LLMs--served via APIs without access to model internals--are especially dominant due to their scalability and ease of deployment. Despite their strong capabilities, these models typically produce generalized responses that overlook personal preferences and reasoning styles. This has led to growing interest in black-box LLM personalization, which aims to tailor model outputs to user-specific context without modifying model parameters. However, existing approaches primarily focus on response-level personalization, attempting to match final outputs without modeling personal thought process. To address this limitation, we propose RPM, a framework for reasoning-level personalization that aligns the model's reasoning process with a user's personalized logic. RPM first constructs statistical user-specific factors by extracting and grouping response-influential features from user history. It then builds personalized reasoning paths that reflect how these factors are used in context. In the inference stage, RPM retrieves reasoning-aligned examples for new queries via feature-level similarity and performs inference conditioned on the structured factors and retrieved reasoning paths, enabling the model to follow user-specific reasoning trajectories. This reasoning-level personalization enhances both predictive accuracy and interpretability by grounding model outputs in user-specific logic through structured information. Extensive experiments across diverse tasks show that RPM consistently outperforms response-level personalization methods, demonstrating the effectiveness of reasoning-level personalization in black-box LLMs. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 12:06:16 GMT"
},
{
"version": "v2",
"created": "Wed, 28 May 2025 13:20:32 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 11:03:01 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Kim",
"Jieyong",
""
],
[
"Kim",
"Tongyoung",
""
],
[
"Yoon",
"Soojin",
""
],
[
"Kim",
"Jaehyung",
""
],
[
"Lee",
"Dongha",
""
]
] |
2505.21124 | Rohit Kanchi | Rohit Sunil Kanchi, Benjamin Melanson, Nithin Somasekharan, Shaowu Pan, Sicheng He | UniFoil: A Universal Dataset of Airfoils in Transitional and Turbulent Regimes for Subsonic and Transonic Flows | null | null | null | null | physics.flu-dyn physics.data-an | http://creativecommons.org/licenses/by-sa/4.0/ | We present UniFoil, a large publicly available universal airfoil dataset based on Reynolds-averaged Navier-Stokes (RANS) simulations. It contains over 500,000 samples spanning a wide range of Reynolds and Mach numbers, capturing both transitional and fully turbulent flows across incompressible to compressible regimes. UniFoil is designed to support machine learning research in fluid dynamics, particularly for modeling complex aerodynamic phenomena. Most existing datasets are limited to incompressible, fully turbulent flows with smooth field characteristics, overlooking the critical physics of laminar\-turbulent transition and shock\-wave interactions\-features that exhibit strong nonlinearity and sharp gradients. UniFoil addresses this limitation by offering a broad spectrum of realistic flow conditions. Turbulent simulations utilize the Spalart\-Allmaras (SA) model, while transitional flows are modeled using an e^N\-based transition prediction method coupled with the SA model. The dataset includes a comprehensive geometry set comprising over 4,800 natural laminar flow (NLF) airfoils and 30,000 fully turbulent (FT) airfoils, covering a diverse range of airfoil designs relevant to aerospace, wind energy, and marine applications. This dataset is also valuable for scientific machine learning, enabling the development of data-driven models that more accurately capture the transport processes associated with laminar-turbulent transition. UniFoil is freely available under a permissive CC\-BY\-SA license. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 12:43:20 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 21:29:56 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Kanchi",
"Rohit Sunil",
""
],
[
"Melanson",
"Benjamin",
""
],
[
"Somasekharan",
"Nithin",
""
],
[
"Pan",
"Shaowu",
""
],
[
"He",
"Sicheng",
""
]
] |
2505.21420 | Hanzhe Liang | Hanzhe Liang | Mentor3AD: Feature Reconstruction-based 3D Anomaly Detection via Multi-modality Mentor Learning | arXiv admin comment: This version has been removed by arXiv administrators as the submitter did not have the rights to agree to the license at the time of submission | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Multimodal feature reconstruction is a promising approach for 3D anomaly detection, leveraging the complementary information from dual modalities. We further advance this paradigm by utilizing multi-modal mentor learning, which fuses intermediate features to further distinguish normal from feature differences. To address these challenges, we propose a novel method called Mentor3AD, which utilizes multi-modal mentor learning. By leveraging the shared features of different modalities, Mentor3AD can extract more effective features and guide feature reconstruction, ultimately improving detection performance. Specifically, Mentor3AD includes a Mentor of Fusion Module (MFM) that merges features extracted from RGB and 3D modalities to create a mentor feature. Additionally, we have designed a Mentor of Guidance Module (MGM) to facilitate cross-modal reconstruction, supported by the mentor feature. Lastly, we introduce a Voting Module (VM) to more accurately generate the final anomaly score. Extensive comparative and ablation studies on MVTec 3D-AD and Eyecandies have verified the effectiveness of the proposed method. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 16:46:28 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Liang",
"Hanzhe",
""
]
] |
2505.21427 | Xianling Mu | Xianling Mu, Joseph Ternasky, Fuat Alican, Yigit Ihlamur | Policy Induction: Predicting Startup Success via Explainable Memory-Augmented In-Context Learning | null | null | null | null | cs.AI cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Early-stage startup investment is a high-risk endeavor characterized by scarce data and uncertain outcomes. Traditional machine learning approaches often require large, labeled datasets and extensive fine-tuning, yet remain opaque and difficult for domain experts to interpret or improve. In this paper, we propose a transparent and data-efficient investment decision framework powered by memory-augmented large language models (LLMs) using in-context learning (ICL). Central to our method is a natural language policy embedded directly into the LLM prompt, enabling the model to apply explicit reasoning patterns and allowing human experts to easily interpret, audit, and iteratively refine the logic. We introduce a lightweight training process that combines few-shot learning with an in-context learning loop, enabling the LLM to update its decision policy iteratively based on structured feedback. With only minimal supervision and no gradient-based optimization, our system predicts startup success far more accurately than existing benchmarks. It is over 20x more precise than random chance, which succeeds 1.9% of the time. It is also 7.1x more precise than the typical 5.6% success rate of top-tier venture capital (VC) firms. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 16:57:07 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 09:16:21 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Mu",
"Xianling",
""
],
[
"Ternasky",
"Joseph",
""
],
[
"Alican",
"Fuat",
""
],
[
"Ihlamur",
"Yigit",
""
]
] |
2505.21649 | Nazia Tasnim | Keanu Nichols, Nazia Tasnim, Yuting Yan, Nicholas Ikechukwu, Elva Zou, Deepti Ghadiyaram, Bryan A. Plummer | Right Side Up? Disentangling Orientation Understanding in MLLMs with Fine-grained Multi-axis Perception Tasks | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Object orientation understanding represents a fundamental challenge in visual perception critical for applications like robotic manipulation and augmented reality. Current vision-language benchmarks fail to isolate this capability, often conflating it with positional relationships and general scene understanding. We introduce DORI (Discriminative Orientation Reasoning Intelligence), a comprehensive benchmark establishing object orientation perception as a primary evaluation target. DORI assesses four dimensions of orientation comprehension: frontal alignment, rotational transformations, relative directional relationships, and canonical orientation understanding. Through carefully curated tasks from 11 datasets spanning 67 object categories across synthetic and real-world scenarios, DORI provides insights on how multi-modal systems understand object orientations. Our evaluation of 15 state-of-the-art vision-language models reveals critical limitations: even the best models achieve only 54.2% accuracy on coarse tasks and 33.0% on granular orientation judgments, with performance deteriorating for tasks requiring reference frame shifts or compound rotations. These findings demonstrate the need for dedicated orientation representation mechanisms, as models show systematic inability to perform precise angular estimations, track orientation changes across viewpoints, and understand compound rotations - suggesting limitations in their internal 3D spatial representations. As the first diagnostic framework specifically designed for orientation awareness in multimodal systems, DORI offers implications for improving robotic control, 3D scene reconstruction, and human-AI interaction in physical environments. DORI data: https://huggingface.co/datasets/appledora/DORI-Benchmark | [
{
"version": "v1",
"created": "Tue, 27 May 2025 18:22:44 GMT"
},
{
"version": "v2",
"created": "Thu, 29 May 2025 13:17:48 GMT"
},
{
"version": "v3",
"created": "Fri, 30 May 2025 03:36:31 GMT"
},
{
"version": "v4",
"created": "Wed, 04 Jun 2025 17:28:29 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Nichols",
"Keanu",
""
],
[
"Tasnim",
"Nazia",
""
],
[
"Yan",
"Yuting",
""
],
[
"Ikechukwu",
"Nicholas",
""
],
[
"Zou",
"Elva",
""
],
[
"Ghadiyaram",
"Deepti",
""
],
[
"Plummer",
"Bryan A.",
""
]
] |
2505.21677 | Emily Wenger | Hung Anh Vu and Galen Reeves and Emily Wenger | What happens when generative AI models train recursively on each others' generated outputs? | 9 pages | null | null | null | cs.LG cs.AI cs.CY | http://creativecommons.org/licenses/by/4.0/ | The internet is full of AI-generated content while also serving as a common source of training data for generative AI (genAI) models. This duality raises the possibility that future genAI models may be trained on other models' generated outputs. Prior work has studied consequences of models training on their own generated outputs, but limited work has considered what happens if models ingest content produced by other models. Given society's increasing dependence on genAI tools, understanding downstream effects of such data-mediated model interactions is critical. To this end, we provide empirical evidence for how data-mediated interactions might unfold in practice, develop a theoretical model for this interactive training process, and show experimentally possible long-term results of such interactions. We find that data-mediated interactions can benefit models by exposing them to novel concepts perhaps missed in original training data, but also can homogenize their performance on shared tasks. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 18:52:34 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 18:34:02 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Vu",
"Hung Anh",
""
],
[
"Reeves",
"Galen",
""
],
[
"Wenger",
"Emily",
""
]
] |
2505.21945 | Maria Kirsanova S. | Maria. S. Kirsanova, Anastasiia A. Farafontova | Deuterated water abundance in the young hot core RCW 120 S2 | Accepted by Astronomy Letters | null | null | null | astro-ph.SR astro-ph.GA | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Since the emission of water molecules cannot be observed from Earth, less abundant isotopologues, such as H$_2^{18}$O and HDO, are used to trace water in star-forming regions. The main aim of this study is to determine HDO abundance in the hot core RCW 120 S2. We performed observations of the hot core in the 200-255~GHz range using the nFLASH230 receiver on the APEX telescope. Two HDO lines were detected toward RCW 120 S2. Their intensities are described by excitation temperature $286\pm2$~K and gas number density $\geq 10^9$~cm$^{-3}$. The emission originates from the hot core rather than the warm dense envelope surrounding a central young stellar object. The HDO column density ranges from $(3.9-7.9)\times10^{13}$~cm$^{-3}$ with a most probable value $5.6\times10^{13}$~cm$^{-3}$. The HDO abundance relative to hydrogen nuclei is $1.5\times10^{-9}$. This HDO abundance value is among the lowest reported for hot cores. Combined with the non-detection of the H$_2^{18}$O line, we conclude that protostellar heating in RCW 120 S2 is still in its early stages. | [
{
"version": "v1",
"created": "Wed, 28 May 2025 03:55:13 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 05:27:08 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Kirsanova",
"Maria. S.",
""
],
[
"Farafontova",
"Anastasiia A.",
""
]
] |
2505.22256 | Yizhong Chen | Yizhong Chen and Zhibin Wang | Improvement of Solid-Fluid Interaction Scheme in Lattice Boltzmann Immiscible Pseudopotential Models | null | null | null | null | physics.flu-dyn | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The pseudopotential model within the Lattice Boltzmann Method (LBM) framework has emerged as a prominent approach in computational fluid dynamics due to its dual strengths in physical intuitiveness and computational tractability. However, when modeling wettability phenomenon, existing solid-fluid interaction schemes exhibit persistent challenges in multi-component immiscible fluid systems, notably manifested through spurious velocity artifacts and unphysical mass-transfer boundary layers. This study presents an improved interaction scheme that preserves implementation simplicity while effectively mitigating these numerical artifacts. Furthermore, leveraging the enhanced isotropy characteristics of eighth-order discrete schemes, we develop a novel boundary treatment methodology addressing second-layer lattice data reconstruction at complex interfaces. To verify the universality of the proposed optimization scheme, four benchmark scenarios, including static contact angle measurement on cylindrical surface, droplet dynamics through confined geometries, immiscible displacement processes, and co-current flow in microchannels, are simulated to demonstrate the proposed scheme's capability. The results show that the improved scheme can well simulate various complex immiscible multiphase flows. | [
{
"version": "v1",
"created": "Wed, 28 May 2025 11:42:05 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 12:42:37 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Chen",
"Yizhong",
""
],
[
"Wang",
"Zhibin",
""
]
] |
2505.22727 | Pavel E. Mancera Pi\~na | Pavel E. Mancera Pi\~na, Justin I. Read, Stacy Kim, Antonino Marasco, Jos\'e A. Benavides, Marcin Glowacki, Gabriele Pezzulli, and Claudia del P. Lagos | The galaxy-halo connection of disc galaxies over six orders of magnitude in stellar mass | A&A, in press. 18 pages (10 figures) + appendices. A couple of minor typos were fixed compared to the previous version | null | null | null | astro-ph.GA astro-ph.CO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | (Abridged) The relations between stellar ($M_\ast$), gas ($M_{\rm gas}$), baryonic ($M_{\rm bar} = M_\ast + M_{\rm gas}$), and dark matter halo mass ($M_{200}$) provide unique constraints on galaxy formation and cosmology. The shape of the relations constrains how galaxies regulate their growth through gas accretion, star formation, and feedback; their scatter probes the stochasticity of galaxy assembly.
Here, we assemble a sample of 49 nearby gas-rich dwarf and massive disc galaxies with unmatched ancillary data. We obtain their gas kinematics and derive their dark matter properties through rotation curve decomposition. Our sample allows us to study the galaxy-halo connection across nearly six orders of magnitude in $M_\ast$. We find that the $M_{\rm gas}-M_{200}$ relation rises monotonically, with galaxies having around 4 per cent of the average cosmological baryon fraction in cold gas. Contrastingly, the $M_\ast-M_{200}$ relation shows a more complex behaviour. A particularly interesting finding is that of a population of baryon-deficient' dwarfs (BDDs) with stellar masses $\sim 1-1.5$ orders of magnitude lower than expected from current models. Yet, baryon-rich galaxies also exist, and we find a large spread in the baryon retention fraction across our galaxies. We compare our findings with semi-analytic and hydrodynamical galaxy formation simulations. While the simulations broadly reproduce most observed features, they struggle to match the BDDs and do not capture the diversity in baryon fractions. Understanding these differences will shed new light on how feedback regulates galaxy formation. Finally, we study the dark matter halo concentration-mass relation. We find that below $M_{200} \sim 10^{11}\,M_\odot$, the concentrations are systematically lower than expected. We discuss whether these results stem from the influence of baryonic physics or the environment. | [
{
"version": "v1",
"created": "Wed, 28 May 2025 18:00:02 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 18:00:05 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Piña",
"Pavel E. Mancera",
""
],
[
"Read",
"Justin I.",
""
],
[
"Kim",
"Stacy",
""
],
[
"Marasco",
"Antonino",
""
],
[
"Benavides",
"José A.",
""
],
[
"Glowacki",
"Marcin",
""
],
[
"Pezzulli",
"Gabriele",
""
],
[
"Lagos",
"Claudia del P.",
""
]
] |
2505.22813 | Josiah Couch | Josiah Couch, Miao Li, Rima Arnaout, Ramy Arnaout | X-Factor: Quality Is a Dataset-Intrinsic Property | 13 pages, 7 figures | null | null | null | cs.LG | http://creativecommons.org/licenses/by-nc-sa/4.0/ | In the universal quest to optimize machine-learning classifiers, three factors -- model architecture, dataset size, and class balance -- have been shown to influence test-time performance but do not fully account for it. Previously, evidence was presented for an additional factor that can be referred to as dataset quality, but it was unclear whether this was actually a joint property of the dataset and the model architecture, or an intrinsic property of the dataset itself. If quality is truly dataset-intrinsic and independent of model architecture, dataset size, and class balance, then the same datasets should perform better (or worse) regardless of these other factors. To test this hypothesis, here we create thousands of datasets, each controlled for size and class balance, and use them to train classifiers with a wide range of architectures, from random forests and support-vector machines to deep networks. We find that classifier performance correlates strongly by subset across architectures ($R^2=0.79$), supporting quality as an intrinsic property of datasets independent of dataset size and class balance and of model architecture. Digging deeper, we find that dataset quality appears to be an emergent property of something more fundamental: the quality of datasets' constituent classes. Thus, quality joins size, class balance, and model architecture as an independent correlate of performance and a separate target for optimizing machine-learning-based classification. | [
{
"version": "v1",
"created": "Wed, 28 May 2025 19:41:37 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 16:02:34 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Couch",
"Josiah",
""
],
[
"Li",
"Miao",
""
],
[
"Arnaout",
"Rima",
""
],
[
"Arnaout",
"Ramy",
""
]
] |
2505.22944 | Angtian Wang | Angtian Wang, Haibin Huang, Jacob Zhiyuan Fang, Yiding Yang, Chongyang Ma | ATI: Any Trajectory Instruction for Controllable Video Generation | null | null | null | null | cs.CV cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We propose a unified framework for motion control in video generation that seamlessly integrates camera movement, object-level translation, and fine-grained local motion using trajectory-based inputs. In contrast to prior methods that address these motion types through separate modules or task-specific designs, our approach offers a cohesive solution by projecting user-defined trajectories into the latent space of pre-trained image-to-video generation models via a lightweight motion injector. Users can specify keypoints and their motion paths to control localized deformations, entire object motion, virtual camera dynamics, or combinations of these. The injected trajectory signals guide the generative process to produce temporally consistent and semantically aligned motion sequences. Our framework demonstrates superior performance across multiple video motion control tasks, including stylized motion effects (e.g., motion brushes), dynamic viewpoint changes, and precise local motion manipulation. Experiments show that our method provides significantly better controllability and visual quality compared to prior approaches and commercial solutions, while remaining broadly compatible with various state-of-the-art video generation backbones. Project page: https://anytraj.github.io/. | [
{
"version": "v1",
"created": "Wed, 28 May 2025 23:49:18 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 23:33:12 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Wang",
"Angtian",
""
],
[
"Huang",
"Haibin",
""
],
[
"Fang",
"Jacob Zhiyuan",
""
],
[
"Yang",
"Yiding",
""
],
[
"Ma",
"Chongyang",
""
]
] |
2505.22990 | PinHan Chen | Pin-Han Chen, Yu-Sheng Lin, Wei-Cheng Lee, Tin-Yu Leu, Po-Hsiang Hsu, Anjana Dissanayake, Sungjin Oh, Chinq-Shiun Chiu | MenTeR: A fully-automated Multi-agenT workflow for end-to-end RF/Analog Circuits Netlist Design | 9 pages, 7 figures, accepted by IEEE ICLAD 2025 | null | null | null | cs.AI cs.ET cs.LG | http://creativecommons.org/licenses/by-nc-sa/4.0/ | RF/Analog design is essential for bridging digital technologies with real-world signals, ensuring the functionality and reliability of a wide range of electronic systems. However, analog design procedures are often intricate, time-consuming and reliant on expert intuition, and hinder the time and cost efficiency of circuit development. To overcome the limitations of the manual circuit design, we introduce MenTeR - a multiagent workflow integrated into an end-to-end analog design framework. By employing multiple specialized AI agents that collaboratively address different aspects of the design process, such as specification understanding, circuit optimization, and test bench validation, MenTeR reduces the dependency on frequent trial-and-error-style intervention. MenTeR not only accelerates the design cycle time but also facilitates a broader exploration of the design space, demonstrating robust capabilities in handling real-world analog systems. We believe that MenTeR lays the groundwork for future "RF/Analog Copilots" that can collaborate seamlessly with human designers. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 01:58:08 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 20:36:51 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Chen",
"Pin-Han",
""
],
[
"Lin",
"Yu-Sheng",
""
],
[
"Lee",
"Wei-Cheng",
""
],
[
"Leu",
"Tin-Yu",
""
],
[
"Hsu",
"Po-Hsiang",
""
],
[
"Dissanayake",
"Anjana",
""
],
[
"Oh",
"Sungjin",
""
],
[
"Chiu",
"Chinq-Shiun",
""
]
] |
2505.23001 | Yize Cheng | Yize Cheng, Wenxiao Wang, Mazda Moayeri, Soheil Feizi | DyePack: Provably Flagging Test Set Contamination in LLMs Using Backdoors | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Open benchmarks are essential for evaluating and advancing large language models, offering reproducibility and transparency. However, their accessibility makes them likely targets of test set contamination. In this work, we introduce DyePack, a framework that leverages backdoor attacks to identify models that used benchmark test sets during training, without requiring access to the loss, logits, or any internal details of the model. Like how banks mix dye packs with their money to mark robbers, DyePack mixes backdoor samples with the test data to flag models that trained on it. We propose a principled design incorporating multiple backdoors with stochastic targets, enabling exact false positive rate (FPR) computation when flagging every model. This provably prevents false accusations while providing strong evidence for every detected case of contamination. We evaluate DyePack on five models across three datasets, covering both multiple-choice and open-ended generation tasks. For multiple-choice questions, it successfully detects all contaminated models with guaranteed FPRs as low as 0.000073% on MMLU-Pro and 0.000017% on Big-Bench-Hard using eight backdoors. For open-ended generation tasks, it generalizes well and identifies all contaminated models on Alpaca with a guaranteed false positive rate of just 0.127% using six backdoors. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 02:22:14 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 15:13:43 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 02:31:16 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Cheng",
"Yize",
""
],
[
"Wang",
"Wenxiao",
""
],
[
"Moayeri",
"Mazda",
""
],
[
"Feizi",
"Soheil",
""
]
] |
2505.23161 | Donato Crisostomi | Antonio D'Orazio, Maria Rosaria Briglia, Donato Crisostomi, Dario Loi, Emanuele Rodol\`a, Iacopo Masi | Implicit Inversion turns CLIP into a Decoder | null | null | null | null | cs.CV cs.AI cs.LG | http://creativecommons.org/licenses/by-nc-sa/4.0/ | CLIP is a discriminative model trained to align images and text in a shared embedding space. Due to its multimodal structure, it serves as the backbone of many generative pipelines, where a decoder is trained to map from the shared space back to images. In this work, we show that image synthesis is nevertheless possible using CLIP alone -- without any decoder, training, or fine-tuning. Our approach optimizes a frequency-aware implicit neural representation that encourages coarse-to-fine generation by stratifying frequencies across network layers. To stabilize this inverse mapping, we introduce adversarially robust initialization, a lightweight Orthogonal Procrustes projection to align local text and image embeddings, and a blending loss that anchors outputs to natural image statistics. Without altering CLIP's weights, this framework unlocks capabilities such as text-to-image generation, style transfer, and image reconstruction. These findings suggest that discriminative models may hold untapped generative potential, hidden in plain sight. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 06:55:26 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 10:30:14 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"D'Orazio",
"Antonio",
""
],
[
"Briglia",
"Maria Rosaria",
""
],
[
"Crisostomi",
"Donato",
""
],
[
"Loi",
"Dario",
""
],
[
"Rodolà",
"Emanuele",
""
],
[
"Masi",
"Iacopo",
""
]
] |
2505.23199 | Si-Wen Li | Si-wen Li, Xun Chen | Chaos in the holographic matrix models for meson and baryon | 44 pages, 14 figures, 1 table, add 2 references to this version | null | null | null | hep-th | http://creativecommons.org/licenses/by/4.0/ | In recent years, the investigation of chaos has become a bridge connecting gravity theory and quantum field theory, especially within the framework of gauge-gravity duality. In this work, we study holographically the chaos in the matrix models for meson and baryon, which are derived from the $\mathrm{D4}/\mathrm{D6}/\overline{\mathrm{D6}}$ approach as a top-down holographic model for QCD. Since these matrix models can be simplified into coupled oscillator models with special parameters, we analyze the chaos in the resultant coupled oscillators. In the analysis of the classical chaos, we calculate numerically the orbits on the Poincar\'e section, the Lyapunov exponent as a function of the total energy and derive the large $N_{c}$ behavior analytically, then discuss the possible phase structure both in the mesonic and baryonic matrix models. These analyses suggest that chaos might serve as an order parameter to detect the gauge theory with spontaneous breaking or restoration of symmetry. Besides, in the analysis of the quantum chaos, we demonstrate the numerical calculation of the OTOCs and analytically derive their large $N_{c}$ behavior by using the perturbation method in quantum mechanics. The numerical calculation illustrates there is a critical temperature, as a critical energy scale, that the OTOC begins to saturate, which covers qualitatively the classical analysis of the Lyapunov exponent. And the large $N_{c}$ analytics indicates the OTOCs are suppressed by the growth of $N_{c}$. Overall, the investigation of chaos in this work may be helpful to identify common features shared by the matrix models, hadronic physics, gauge theory, quantum mechanics, and gravity theory. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 07:35:53 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 00:49:20 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Li",
"Si-wen",
""
],
[
"Chen",
"Xun",
""
]
] |
2505.23276 | Maged S. Al-Shaibani | Maged S. Al-Shaibani, Moataz Ahmed | The Arabic AI Fingerprint: Stylometric Analysis and Detection of Large Language Models Text | null | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Large Language Models (LLMs) have achieved unprecedented capabilities in generating human-like text, posing subtle yet significant challenges for information integrity across critical domains, including education, social media, and academia, enabling sophisticated misinformation campaigns, compromising healthcare guidance, and facilitating targeted propaganda. This challenge becomes severe, particularly in under-explored and low-resource languages like Arabic. This paper presents a comprehensive investigation of Arabic machine-generated text, examining multiple generation strategies (generation from the title only, content-aware generation, and text refinement) across diverse model architectures (ALLaM, Jais, Llama, and GPT-4) in academic, and social media domains. Our stylometric analysis reveals distinctive linguistic patterns differentiating human-written from machine-generated Arabic text across these varied contexts. Despite their human-like qualities, we demonstrate that LLMs produce detectable signatures in their Arabic outputs, with domain-specific characteristics that vary significantly between different contexts. Based on these insights, we developed BERT-based detection models that achieved exceptional performance in formal contexts (up to 99.9\% F1-score) with strong precision across model architectures. Our cross-domain analysis confirms generalization challenges previously reported in the literature. To the best of our knowledge, this work represents the most comprehensive investigation of Arabic machine-generated text to date, uniquely combining multiple prompt generation methods, diverse model architectures, and in-depth stylometric analysis across varied textual domains, establishing a foundation for developing robust, linguistically-informed detection systems essential for preserving information integrity in Arabic-language contexts. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 09:24:00 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 15:16:04 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Al-Shaibani",
"Maged S.",
""
],
[
"Ahmed",
"Moataz",
""
]
] |
2505.23455 | Narmin Nasibova | Narmin Nasibova and Xerxes D. Arsiwalla | Pion Phenomenology from the Thermal Soft-Wall Model of Holographic QCD | 34 pages, 27 figures | null | null | null | hep-ph | http://creativecommons.org/licenses/by/4.0/ | We investigate the temperature dependence of the electromagnetic form factor of the pion, its generalized parton distribution $H_{\pi}(T)$, effective mass $m_{\pi}(T)$, radius $r_{\pi}(T)$, electromagnetic charge density $\rho_{\pi}(b, T)$, pion-nucleon coupling constant $g_{\pi NN}(T)$, and the pion-$\Delta$ coupling $g_{\pi \Delta \Delta}(T)$ within the thermal soft-wall model of AdS/QCD. The thermal pion form factor is extrapolated from the zero-temperature case, and based on this, the generalized parton distribution, charge density, and radius are studied at finite temperature. These quantities were analyzed using a thermal dilaton field in the five-dimensional action. Moreover, we determine the theoretical expression for the temperature-dependent pion-nucleon, as well as pion-$\Delta$ coupling, using thermal profile functions of nucleons, $\Delta$ baryons, and the pion. Our results show that the values of these quantities decrease with increasing temperature and vanish near the critical temperature. Our results reproduce expected features of low-energy hadron dynamics, thus validating the phenomenological utility of the thermal soft-wall model. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 14:01:34 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 14:25:20 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Nasibova",
"Narmin",
""
],
[
"Arsiwalla",
"Xerxes D.",
""
]
] |
2505.23470 | Chenjie Li | Chenjie Li, Amir Gilad, Boris Glavic, Zhengjie Miao, and Sudeepa Roy | Refining Labeling Functions with Limited Labeled Data | techreport | null | null | null | cs.LG cs.IT math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Programmatic weak supervision (PWS) significantly reduces human effort for labeling data by combining the outputs of user-provided labeling functions (LFs) on unlabeled datapoints. However, the quality of the generated labels depends directly on the accuracy of the LFs. In this work, we study the problem of fixing LFs based on a small set of labeled examples. Towards this goal, we develop novel techniques for repairing a set of LFs by minimally changing their results on the labeled examples such that the fixed LFs ensure that (i) there is sufficient evidence for the correct label of each labeled datapoint and (ii) the accuracy of each repaired LF is sufficiently high. We model LFs as conditional rules which enables us to refine them, i.e., to selectively change their output for some inputs. We demonstrate experimentally that our system improves the quality of LFs based on surprisingly small sets of labeled datapoints. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 14:26:11 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 04:14:44 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Li",
"Chenjie",
""
],
[
"Gilad",
"Amir",
""
],
[
"Glavic",
"Boris",
""
],
[
"Miao",
"Zhengjie",
""
],
[
"Roy",
"Sudeepa",
""
]
] |
2505.23527 | Raj Ghugare | Raj Ghugare and Benjamin Eysenbach | Normalizing Flows are Capable Models for RL | Project page with code - https://rajghugare19.github.io/nf4rl/ | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Modern reinforcement learning (RL) algorithms have found success by using powerful probabilistic models, such as transformers, energy-based models, and diffusion/flow-based models. To this end, RL researchers often choose to pay the price of accommodating these models into their algorithms -- diffusion models are expressive, but are computationally intensive due to their reliance on solving differential equations, while autoregressive transformer models are scalable but typically require learning discrete representations. Normalizing flows (NFs), by contrast, seem to provide an appealing alternative, as they enable likelihoods and sampling without solving differential equations or autoregressive architectures. However, their potential in RL has received limited attention, partly due to the prevailing belief that normalizing flows lack sufficient expressivity. We show that this is not the case. Building on recent work in NFs, we propose a single NF architecture which integrates seamlessly into RL algorithms, serving as a policy, Q-function, and occupancy measure. Our approach leads to much simpler algorithms, and achieves higher performance in imitation learning, offline, goal conditioned RL and unsupervised RL. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 15:06:22 GMT"
},
{
"version": "v2",
"created": "Fri, 30 May 2025 18:47:02 GMT"
},
{
"version": "v3",
"created": "Tue, 03 Jun 2025 18:43:14 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Ghugare",
"Raj",
""
],
[
"Eysenbach",
"Benjamin",
""
]
] |
2505.23585 | Li Dong | Yaru Hao, Li Dong, Xun Wu, Shaohan Huang, Zewen Chi, Furu Wei | On-Policy RL with Optimal Reward Baseline | null | null | null | null | cs.LG cs.CL | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Reinforcement learning algorithms are fundamental to align large language models with human preferences and to enhance their reasoning capabilities. However, current reinforcement learning algorithms often suffer from training instability due to loose on-policy constraints and computational inefficiency due to auxiliary models. In this work, we propose On-Policy RL with Optimal reward baseline (OPO), a novel and simplified reinforcement learning algorithm designed to address these challenges. OPO emphasizes the importance of exact on-policy training, which empirically stabilizes the training process and enhances exploration. Moreover, OPO integrates a practically feasible formulation of the optimal reward baseline that minimizes gradient variance. We evaluate OPO on mathematical reasoning benchmarks. The results demonstrate its superior performance and training stability without additional models or regularization terms. Furthermore, OPO achieves lower policy shifts and higher output entropy, encouraging more diverse and less repetitive responses. These results highlight OPO as a promising direction for stable and effective reinforcement learning in large language model alignment and reasoning tasks. The implementation is merged into the verl library at https://verl.readthedocs.io/en/latest/algo/opo.html. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 15:58:04 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 01:41:37 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Hao",
"Yaru",
""
],
[
"Dong",
"Li",
""
],
[
"Wu",
"Xun",
""
],
[
"Huang",
"Shaohan",
""
],
[
"Chi",
"Zewen",
""
],
[
"Wei",
"Furu",
""
]
] |
2505.23633 | Long Zhang | Xiao-Dong Lin, Long Zhang | Measuring topological invariants of even-dimensional non-Hermitian systems through quench dynamics | 10 pages, 4 figures; Figures and references are updated; complementary to arXiv: 2410.13241 | null | null | null | cond-mat.quant-gas cond-mat.mes-hall quant-ph | http://creativecommons.org/licenses/by/4.0/ | The accurate determination of non-Hermitian (NH) topological invariants plays a central role in the study of NH topological phases. In this work, we propose a general framework for directly measuring NH topological invariants in even-dimensional systems through quench dynamics. Our approach hinges on constructing an auxiliary Hermitian matrix topologically equivalent to the original NH Hamiltonian, enabling topological characterization via reduced-dimensional momentum subspaces called band inversion surfaces (BISs). A key insight lies in the emergence of chiral symmetry in the NH Hamiltonian specifically on BISs -- a critical property that allows extension of the dynamical characterization scheme previously developed for odd-dimensional NH systems with chiral or sublattice symmetry [Lin et al., Phys. Rev. Res. 7, L012060 (2025)]. We show that NH topological invariants can be extracted from the winding patterns of a dynamical field constructed from post-quench spin textures on BISs. We demonstrate our approach through a detailed analysis of NH Chern insulators and then extend the framework to higher even-dimensional systems by introducing second-order BISs for characterization. This work establishes an experimentally accessible protocol for detecting NH topological invariants in quantum platforms. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 16:41:09 GMT"
},
{
"version": "v2",
"created": "Mon, 02 Jun 2025 09:48:03 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 14:12:11 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Lin",
"Xiao-Dong",
""
],
[
"Zhang",
"Long",
""
]
] |
2505.23637 | Dashti Ali | Dashti A. Ali, Richard K. G. Do, William R. Jarnagin, Aras T. Asaad, Amber L. Simpson | Comparing the Effects of Persistence Barcodes Aggregation and Feature Concatenation on Medical Imaging | 16 pages, 8 figures | null | null | null | cs.CV cs.AI | http://creativecommons.org/licenses/by/4.0/ | In medical image analysis, feature engineering plays an important role in the design and performance of machine learning models. Persistent homology (PH), from the field of topological data analysis (TDA), demonstrates robustness and stability to data perturbations and addresses the limitation from traditional feature extraction approaches where a small change in input results in a large change in feature representation. Using PH, we store persistent topological and geometrical features in the form of the persistence barcode whereby large bars represent global topological features and small bars encapsulate geometrical information of the data. When multiple barcodes are computed from 2D or 3D medical images, two approaches can be used to construct the final topological feature vector in each dimension: aggregating persistence barcodes followed by featurization or concatenating topological feature vectors derived from each barcode. In this study, we conduct a comprehensive analysis across diverse medical imaging datasets to compare the effects of the two aforementioned approaches on the performance of classification models. The results of this analysis indicate that feature concatenation preserves detailed topological information from individual barcodes, yields better classification performance and is therefore a preferred approach when conducting similar experiments. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 16:45:33 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 14:50:55 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Ali",
"Dashti A.",
""
],
[
"Do",
"Richard K. G.",
""
],
[
"Jarnagin",
"William R.",
""
],
[
"Asaad",
"Aras T.",
""
],
[
"Simpson",
"Amber L.",
""
]
] |
2505.23703 | Ruida Wang | Ruida Wang, Yuxin Li, Yi R. Fung, Tong Zhang | Let's Reason Formally: Natural-Formal Hybrid Reasoning Enhances LLM's Math Capability | null | null | null | null | cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Enhancing the mathematical reasoning capabilities of LLMs has garnered significant attention in both the mathematical and computer science communities. Recent works have made substantial progress in both Natural Language (NL) reasoning and Formal Language (FL) reasoning by leveraging the potential of pure Reinforcement Learning (RL) methods on base models. However, RL approaches struggle to impart new capabilities not presented in the base model, highlighting the need to integrate more knowledge like FL into NL math reasoning effectively. Yet, this integration is challenging due to inherent disparities in problem structure and reasoning format between NL and FL. To address these challenges, we introduce **NL-FL HybridReasoning**, an end-to-end framework designed to incorporate the FL expert into NL math problem-solving. To bridge the NL and FL input format gap, we propose the *NL-FL Problem Alignment* method, which reformulates the Question-Answering (QA) problems in NL as existence theorems in FL. Subsequently, the *Mixed Problem Input* technique we provide enables the FL reasoner to handle both QA and existence problems concurrently. Lastly, we mitigate the NL and FL output format gap in reasoning through an LLM-based *Answer Extraction* mechanism. Comprehensive experiments demonstrate that the **HybridReasoning** framework achieves **89.80%** and **84.34%** accuracy rates on the MATH-500 and the AMC benchmarks, surpassing the NL baseline by 4.60% and 4.82%, respectively. Notably, some problems resolved by our framework remain unsolved by the NL baseline model even under a larger number of trials. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 17:39:30 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 13:18:59 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Wang",
"Ruida",
""
],
[
"Li",
"Yuxin",
""
],
[
"Fung",
"Yi R.",
""
],
[
"Zhang",
"Tong",
""
]
] |
2505.23786 | Kazuki Egashira | Kazuki Egashira, Robin Staab, Mark Vero, Jingxuan He, Martin Vechev | Mind the Gap: A Practical Attack on GGUF Quantization | ICML 2025 | null | null | null | cs.CR cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | With the increasing size of frontier LLMs, post-training quantization has become the standard for memory-efficient deployment. Recent work has shown that basic rounding-based quantization schemes pose security risks, as they can be exploited to inject malicious behaviors into quantized models that remain hidden in full precision. However, existing attacks cannot be applied to more complex quantization methods, such as the GGUF family used in the popular ollama and llama$.$cpp frameworks. In this work, we address this gap by introducing the first attack on GGUF. Our key insight is that the quantization error -- the difference between the full-precision weights and their (de-)quantized version -- provides sufficient flexibility to construct malicious quantized models that appear benign in full precision. Leveraging this, we develop an attack that trains the target malicious LLM while constraining its weights based on quantization errors. We demonstrate the effectiveness of our attack on three popular LLMs across nine GGUF quantization data types on three diverse attack scenarios: insecure code generation ($\Delta$=$88.7\%$), targeted content injection ($\Delta$=$85.0\%$), and benign instruction refusal ($\Delta$=$30.1\%$). Our attack highlights that (1) the most widely used post-training quantization method is susceptible to adversarial interferences, and (2) the complexity of quantization schemes alone is insufficient as a defense. | [
{
"version": "v1",
"created": "Sat, 24 May 2025 16:30:37 GMT"
},
{
"version": "v2",
"created": "Mon, 02 Jun 2025 16:21:07 GMT"
},
{
"version": "v3",
"created": "Tue, 03 Jun 2025 19:21:57 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Egashira",
"Kazuki",
""
],
[
"Staab",
"Robin",
""
],
[
"Vero",
"Mark",
""
],
[
"He",
"Jingxuan",
""
],
[
"Vechev",
"Martin",
""
]
] |
2505.23811 | Hadi Askari | Hadi Askari, Shivanshu Gupta, Fei Wang, Anshuman Chhabra, Muhao Chen | LayerIF: Estimating Layer Quality for Large Language Models using Influence Functions | Under Review | null | null | null | cs.CL cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Pretrained Large Language Models (LLMs) achieve strong performance across a wide range of tasks, yet exhibit substantial variability in the various layers' training quality with respect to specific downstream applications, limiting their downstream performance. It is therefore critical to estimate layer-wise training quality in a manner that accounts for both model architecture and training data. However, existing approaches predominantly rely on model-centric heuristics (such as spectral statistics, outlier detection, or uniform allocation) while overlooking the influence of data. To address these limitations, we propose LayerIF, a data-driven framework that leverages Influence Functions to quantify the training quality of individual layers in a principled and task-sensitive manner. By isolating each layer's gradients and measuring the sensitivity of the validation loss to training examples by computing layer-wise influences, we derive data-driven estimates of layer importance. Notably, our method produces task-specific layer importance estimates for the same LLM, revealing how layers specialize for different test-time evaluation tasks. We demonstrate the utility of our scores by leveraging them for two downstream applications: (a) expert allocation in LoRA-MoE architectures and (b) layer-wise sparsity distribution for LLM pruning. Experiments across multiple LLM architectures demonstrate that our model-agnostic, influence-guided allocation leads to consistent gains in task performance. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 10:55:12 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 21:18:10 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Askari",
"Hadi",
""
],
[
"Gupta",
"Shivanshu",
""
],
[
"Wang",
"Fei",
""
],
[
"Chhabra",
"Anshuman",
""
],
[
"Chen",
"Muhao",
""
]
] |
2505.23891 | Panos Betzios | Panos Betzios | A microscopic Normal Matrix Model for $(A)dS_2$ | 61 pages, 10 figures, v2:corrected typos and errors in appendices | null | null | null | hep-th math-ph math.MP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We describe the duality between the gravitating $c=1$ (compact) Sine-Gordon model and a normal matrix model. From a two-dimensional quantum gravity perspective and due to the periodic nature of the potential, this model admits both anti-de Sitter and de-Sitter saddles, similarly to simpler models of Sine-Dilaton gravity, as well as more complicated interpolating "wineglass wormhole" geometries. From a string theory perspective the Euclidean de-Sitter (genus zero) saddles are related to the presence of a classical entropic contribution associated to the target space geometry. The gravitating Sine-Gordon model corresponds to a well defined CFT by construction and the eigenvalues of the dual normal matrix model are supported in a compact region of the complex plane. The duality with the normal matrix model is operationally defined even for a finite, but sufficiently large matrix size $N$, depending on the precise observable to be determined. We define and study a "microscopic" version of the large-N limit that allows us to recover non-perturbative results for all physical observables. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 18:00:00 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 20:27:52 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Betzios",
"Panos",
""
]
] |
2505.24073 | Chan-Wei Hu | Chan-Wei Hu, Yueqi Wang, Shuo Xing, Chia-Ju Chen, Zhengzhong Tu | mRAG: Elucidating the Design Space of Multi-modal Retrieval-Augmented Generation | 16 pages, 11 figures | null | null | null | cs.AI cs.CL cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large Vision-Language Models (LVLMs) have made remarkable strides in multimodal tasks such as visual question answering, visual grounding, and complex reasoning. However, they remain limited by static training data, susceptibility to hallucinations, and inability to verify claims against up-to-date, external evidence, compromising their performance in dynamic real-world applications. Retrieval-Augmented Generation (RAG) offers a practical solution to mitigate these challenges by allowing the LVLMs to access large-scale knowledge databases via retrieval mechanisms, thereby grounding model outputs in factual, contextually relevant information. Here in this paper, we conduct the first systematic dissection of the multimodal RAG pipeline for LVLMs, explicitly investigating (1) the retrieval phase: on the modality configurations and retrieval strategies, (2) the re-ranking stage: on strategies to mitigate positional biases and improve the relevance of retrieved evidence, and (3) the generation phase: we further investigate how to best integrate retrieved candidates into the final generation process. Finally, we extend to explore a unified agentic framework that integrates re-ranking and generation through self-reflection, enabling LVLMs to select relevant evidence and suppress irrelevant context dynamically. Our full-stack exploration of RAG for LVLMs yields substantial insights, resulting in an average performance boost of 5% without any fine-tuning. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 23:32:03 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Hu",
"Chan-Wei",
""
],
[
"Wang",
"Yueqi",
""
],
[
"Xing",
"Shuo",
""
],
[
"Chen",
"Chia-Ju",
""
],
[
"Tu",
"Zhengzhong",
""
]
] |
2505.24075 | Alec Linot | Alec J. Linot, Barbara Lopez-Doriga, Yonghong Zhong, and Kunihiko Taira | Extracting dominant dynamics about unsteady base flows | null | null | null | null | physics.flu-dyn | http://creativecommons.org/licenses/by/4.0/ | A wide range of techniques exist for extracting the dominant flow dynamics and features about steady, or periodic base flows. However, there have been limited efforts in extracting the dominant dynamics about unsteady, aperiodic base flow. These flows appear in many applications such as when there is a sudden change in the flow rate through a pipe, when an airfoil experiences stall, or when a vortex forms. For these unsteady flows, it is valuable to know not only the dynamics of the base flow but also the features that form around this base flow. Here, we discuss the current state of research on extracting important flow structures and their dynamics in such cases with time-varying base flows. In particular, we consider data-driven decompositions, operator-based methods, causality analysis, and some other approaches. We also offer an outlook and call attention to key areas that require future efforts. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 23:40:50 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 18:15:39 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Linot",
"Alec J.",
""
],
[
"Lopez-Doriga",
"Barbara",
""
],
[
"Zhong",
"Yonghong",
""
],
[
"Taira",
"Kunihiko",
""
]
] |
2505.24082 | Jonathan Katz | J. I. Katz | Collimation of Fast Radio Burster 20201124A; Repeaters vs. Apparent Non-Repeaters | 8 pp Added reference | RNAAS 9, 83 (2025) (preliminary and incomplete) | null | null | astro-ph.HE | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The recent report of a period in the active repeating Fast Radio Burster 20201124A and of its spindown rate place bounds on the solid angle of its emission on the basis of energetics. The bound depends on the (unknown) efficiency of conversion of rotational energy to coherent radio emission and implies a lower bound on the Lorentz factor of the radiating charges. Bursts may be emitted along the magnetic dipole axis, in repeaters aligned with the rotational axis and the line of sight but misaligned in apparent non-repeaters. This may explain the difficulty of finding periodicity in repeaters and the low duty cycle of apparent non-repeaters. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 00:03:32 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 15:34:42 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Katz",
"J. I.",
""
]
] |
2505.24198 | Yitang Li | Yitang Li, Yuanhang Zhang, Wenli Xiao, Chaoyi Pan, Haoyang Weng, Guanqi He, Tairan He, Guanya Shi | Hold My Beer: Learning Gentle Humanoid Locomotion and End-Effector Stabilization Control | null | null | null | null | cs.RO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Can your humanoid walk up and hand you a full cup of beer, without spilling a drop? While humanoids are increasingly featured in flashy demos like dancing, delivering packages, traversing rough terrain, fine-grained control during locomotion remains a significant challenge. In particular, stabilizing a filled end-effector (EE) while walking is far from solved, due to a fundamental mismatch in task dynamics: locomotion demands slow-timescale, robust control, whereas EE stabilization requires rapid, high-precision corrections. To address this, we propose SoFTA, a Slow-Fast Two-Agent framework that decouples upper-body and lower-body control into separate agents operating at different frequencies and with distinct rewards. This temporal and objective separation mitigates policy interference and enables coordinated whole-body behavior. SoFTA executes upper-body actions at 100 Hz for precise EE control and lower-body actions at 50 Hz for robust gait. It reduces EE acceleration by 2-5x relative to baselines and performs much closer to human-level stability, enabling delicate tasks such as carrying nearly full cups, capturing steady video during locomotion, and disturbance rejection with EE stability. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 04:18:09 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 22:45:46 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Li",
"Yitang",
""
],
[
"Zhang",
"Yuanhang",
""
],
[
"Xiao",
"Wenli",
""
],
[
"Pan",
"Chaoyi",
""
],
[
"Weng",
"Haoyang",
""
],
[
"He",
"Guanqi",
""
],
[
"He",
"Tairan",
""
],
[
"Shi",
"Guanya",
""
]
] |
2505.24293 | James Golden | James R. Golden | Large Language Models are Locally Linear Mappings | minor updates to Fig 2; code available at https://github.com/jamesgolden1/llms-are-llms | null | null | null | cs.LG cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | We demonstrate that the inference operations of several open-weight large language models (LLMs) can be mapped to an exactly equivalent linear system for an input sequence without modifying the model weights or altering output predictions. Extending techniques from image diffusion models that exhibit local or piecewise linearity, we strategically alter the gradient computation with respect to a given input sequence for a next-token prediction such that the Jacobian of the model nearly exactly reproduces the forward prediction with a linear system. We demonstrate this approach across models (Llama 3, Gemma 3, Qwen 3, Phi 4, Mistral Ministral and OLMo 2, up to Llama 3.3 70B Q4) and show through the singular value decomposition of the detached Jacobian that these LLMs operate in extremely low-dimensional subspaces where many of the largest singular vectors decode to concepts related to the most-likely output token. This approach also allows us to examine the operation of each successive layer (and its attention and MLP components) as nearly-exact linear systems and observe the emergence of semantic concepts. Despite their expressive power and global nonlinearity, modern LLMs can be interpreted through nearly-exact locally linear decompositions that provide insights into their internal representations and reveal interpretable semantic structures in the next-token prediction process. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 07:08:33 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 03:50:57 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Golden",
"James R.",
""
]
] |
2505.24298 | Wei Fu | Wei Fu, Jiaxuan Gao, Xujie Shen, Chen Zhu, Zhiyu Mei, Chuyi He, Shusheng Xu, Guo Wei, Jun Mei, Jiashu Wang, Tongkai Yang, Binhang Yuan, Yi Wu | AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Reinforcement learning (RL) has become a dominant paradigm for training large language models (LLMs), particularly for reasoning tasks. Effective RL for LLMs requires massive parallelization and poses an urgent need for efficient training systems. Most existing large-scale RL systems for LLMs are synchronous, alternating generation and training in a batch setting where rollouts in each training batch are generated by the same model. This approach stabilizes RL training but suffers from severe system-level inefficiency: generation must wait until the longest output in the batch is completed before model updates, resulting in GPU underutilization. We present AReaL, a fully asynchronous RL system that completely decouples generation from training. Rollout workers in AReaL continuously generate new outputs without waiting, while training workers update the model whenever a batch of data is collected. AReaL also incorporates a collection of system-level optimizations, leading to substantially higher GPU utilization. To stabilize RL training, AReaL balances the workload of rollout and training workers to control data staleness, and adopts a staleness-enhanced PPO variant to better handle outdated training samples. Extensive experiments on math and code reasoning benchmarks show that AReaL achieves up to 2.77$\times$ training speedup compared to synchronous systems with the same number of GPUs and matched or improved final performance. The code of AReaL is available at https://github.com/inclusionAI/AReaL/. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 07:18:25 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 11:42:19 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Fu",
"Wei",
""
],
[
"Gao",
"Jiaxuan",
""
],
[
"Shen",
"Xujie",
""
],
[
"Zhu",
"Chen",
""
],
[
"Mei",
"Zhiyu",
""
],
[
"He",
"Chuyi",
""
],
[
"Xu",
"Shusheng",
""
],
[
"Wei",
"Guo",
""
],
[
"Mei",
"Jun",
""
],
[
"Wang",
"Jiashu",
""
],
[
"Yang",
"Tongkai",
""
],
[
"Yuan",
"Binhang",
""
],
[
"Wu",
"Yi",
""
]
] |
2505.24371 | Fadoua Ghourabi | Md Intisar Chowdhury, Kittinun Aukkapinyo, Hiroshi Fujimura, Joo Ann Woo, Wasu Wasusatein, Fadoua Ghourabi | Grid-LOGAT: Grid Based Local and Global Area Transcription for Video Question Answering | Copyright 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works | null | null | null | cs.CV cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper, we propose a Grid-based Local and Global Area Transcription (Grid-LoGAT) system for Video Question Answering (VideoQA). The system operates in two phases. First, extracting text transcripts from video frames using a Vision-Language Model (VLM). Next, processing questions using these transcripts to generate answers through a Large Language Model (LLM). This design ensures image privacy by deploying the VLM on edge devices and the LLM in the cloud. To improve transcript quality, we propose grid-based visual prompting, which extracts intricate local details from each grid cell and integrates them with global information. Evaluation results show that Grid-LoGAT, using the open-source VLM (LLaVA-1.6-7B) and LLM (Llama-3.1-8B), outperforms state-of-the-art methods with similar baseline models on NExT-QA and STAR-QA datasets with an accuracy of 65.9% and 50.11% respectively. Additionally, our method surpasses the non-grid version by 24 points on localization-based questions we created using NExT-QA. (This paper is accepted by IEEE ICIP 2025.) | [
{
"version": "v1",
"created": "Fri, 30 May 2025 09:04:30 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 06:42:06 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Chowdhury",
"Md Intisar",
""
],
[
"Aukkapinyo",
"Kittinun",
""
],
[
"Fujimura",
"Hiroshi",
""
],
[
"Woo",
"Joo Ann",
""
],
[
"Wasusatein",
"Wasu",
""
],
[
"Ghourabi",
"Fadoua",
""
]
] |
2505.24434 | Md Shahriar Rahim Siddiqui | Md Shahriar Rahim Siddiqui, Moshe Eliasof, Eldad Haber | Graph Flow Matching: Enhancing Image Generation with Neighbor-Aware Flow Fields | null | null | null | null | cs.LG cs.CV | http://creativecommons.org/licenses/by/4.0/ | Flow matching casts sample generation as learning a continuous-time velocity field that transports noise to data. Existing flow matching networks typically predict each point's velocity independently, considering only its location and time along its flow trajectory, and ignoring neighboring points. However, this pointwise approach may overlook correlations between points along the generation trajectory that could enhance velocity predictions, thereby improving downstream generation quality. To address this, we propose Graph Flow Matching (GFM), a lightweight enhancement that decomposes the learned velocity into a reaction term -- any standard flow matching network -- and a diffusion term that aggregates neighbor information via a graph neural module. This reaction-diffusion formulation retains the scalability of deep flow models while enriching velocity predictions with local context, all at minimal additional computational cost. Operating in the latent space of a pretrained variational autoencoder, GFM consistently improves Fr\'echet Inception Distance (FID) and recall across five image generation benchmarks (LSUN Church, LSUN Bedroom, FFHQ, AFHQ-Cat, and CelebA-HQ at $256\times256$), demonstrating its effectiveness as a modular enhancement to existing flow matching architectures. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 10:17:50 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 08:33:25 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Siddiqui",
"Md Shahriar Rahim",
""
],
[
"Eliasof",
"Moshe",
""
],
[
"Haber",
"Eldad",
""
]
] |
2505.24492 | David Steinmann | David Steinmann, Wolfgang Stammer, Antonia W\"ust, Kristian Kersting | Object Centric Concept Bottlenecks | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Developing high-performing, yet interpretable models remains a critical challenge in modern AI. Concept-based models (CBMs) attempt to address this by extracting human-understandable concepts from a global encoding (e.g., image encoding) and then applying a linear classifier on the resulting concept activations, enabling transparent decision-making. However, their reliance on holistic image encodings limits their expressiveness in object-centric real-world settings and thus hinders their ability to solve complex vision tasks beyond single-label classification. To tackle these challenges, we introduce Object-Centric Concept Bottlenecks (OCB), a framework that combines the strengths of CBMs and pre-trained object-centric foundation models, boosting performance and interpretability. We evaluate OCB on complex image datasets and conduct a comprehensive ablation study to analyze key components of the framework, such as strategies for aggregating object-concept encodings. The results show that OCB outperforms traditional CBMs and allows one to make interpretable decisions for complex visual tasks. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 11:45:05 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 06:54:15 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Steinmann",
"David",
""
],
[
"Stammer",
"Wolfgang",
""
],
[
"Wüst",
"Antonia",
""
],
[
"Kersting",
"Kristian",
""
]
] |
2505.24554 | Anna Sofia Lippolis | Anna Sofia Lippolis, Minh Davide Ragagni, Paolo Ciancarini, Andrea Giovanni Nuzzolese, Valentina Presutti | Bench4KE: Benchmarking Automated Competency Question Generation | null | null | null | null | cs.CL cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The availability of Large Language Models (LLMs) presents a unique opportunity to reinvigorate research on Knowledge Engineering (KE) automation, a trend already evident in recent efforts developing LLM-based methods and tools for the automatic generation of Competency Questions (CQs). However, the evaluation of these tools lacks standardisation. This undermines the methodological rigour and hinders the replication and comparison of results. To address this gap, we introduce Bench4KE, an extensible API-based benchmarking system for KE automation. Its first release focuses on evaluating tools that generate CQs automatically. CQs are natural language questions used by ontology engineers to define the functional requirements of an ontology. Bench4KE provides a curated gold standard consisting of CQ datasets from four real-world ontology projects. It uses a suite of similarity metrics to assess the quality of the CQs generated. We present a comparative analysis of four recent CQ generation systems, which are based on LLMs, establishing a baseline for future research. Bench4KE is also designed to accommodate additional KE automation tasks, such as SPARQL query generation, ontology testing and drafting. Code and datasets are publicly available under the Apache 2.0 license. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 13:03:42 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 09:08:05 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Lippolis",
"Anna Sofia",
""
],
[
"Ragagni",
"Minh Davide",
""
],
[
"Ciancarini",
"Paolo",
""
],
[
"Nuzzolese",
"Andrea Giovanni",
""
],
[
"Presutti",
"Valentina",
""
]
] |
2505.24603 | Omri Lev | Omri Lev, Vishwak Srinivasan, Moshe Shenfeld, Katrina Ligett, Ayush Sekhari, Ashia C. Wilson | The Gaussian Mixing Mechanism: Renyi Differential Privacy via Gaussian Sketches | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Gaussian sketching, which consists of pre-multiplying the data with a random Gaussian matrix, is a widely used technique for multiple problems in data science and machine learning, with applications spanning computationally efficient optimization, coded computing, and federated learning. This operation also provides differential privacy guarantees due to its inherent randomness. In this work, we revisit this operation through the lens of Renyi Differential Privacy (RDP), providing a refined privacy analysis that yields significantly tighter bounds than prior results. We then demonstrate how this improved analysis leads to performance improvement in different linear regression settings, establishing theoretical utility guarantees. Empirically, our methods improve performance across multiple datasets and, in several cases, reduce runtime. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 13:52:48 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 16:02:22 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Lev",
"Omri",
""
],
[
"Srinivasan",
"Vishwak",
""
],
[
"Shenfeld",
"Moshe",
""
],
[
"Ligett",
"Katrina",
""
],
[
"Sekhari",
"Ayush",
""
],
[
"Wilson",
"Ashia C.",
""
]
] |
2505.24688 | Qinglin Zhu | Qinglin Zhu, Runcong Zhao, Hanqi Yan, Yulan He, Yudong Chen, Lin Gui | Soft Reasoning: Navigating Solution Spaces in Large Language Models through Controlled Embedding Exploration | Accepted as a Spotlight at ICML 2025 | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large Language Models (LLMs) struggle with complex reasoning due to limited diversity and inefficient search. We propose Soft Reasoning, an embedding-based search framework that optimises the embedding of the first token to guide generation. It combines (1) embedding perturbation for controlled exploration and (2) Bayesian optimisation to refine embeddings via a verifier-guided objective, balancing exploration and exploitation. This approach improves reasoning accuracy and coherence while avoiding reliance on heuristic search. Experiments demonstrate superior correctness with minimal computation, making it a scalable, model-agnostic solution. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 15:11:52 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 08:11:18 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Zhu",
"Qinglin",
""
],
[
"Zhao",
"Runcong",
""
],
[
"Yan",
"Hanqi",
""
],
[
"He",
"Yulan",
""
],
[
"Chen",
"Yudong",
""
],
[
"Gui",
"Lin",
""
]
] |
2505.24772 | Chen Wei | S. H. Curnoe, D. Gajera, and C. Wei | Mixed state concurrence for symmetric systems | null | null | null | null | quant-ph cond-mat.other | http://creativecommons.org/licenses/by/4.0/ | We present a method to quantify entanglement in mixed states of highly symmetric systems. Symmetry constrains interactions between parts and predicts the degeneracies of the states. While symmetry alone produces entangled eigenstates, the thermal mixed state (density) which contains all of the eigenstate densities weighted by their Boltzmann factors is not necessarily as entangled as the eigenstates themselves because generally the mixed state can be re-expressed as a sum over densities which are less entangled. The entanglement of the mixed state is the minimum obtained by considering all such re-expressions, but there is no well-defined approach to solving this problem generally. Our method uses symmetry to explicitly construct unentangled densities, which are then optimally included in the thermal mixed state, resulting in a quantitative measure of entanglement that accounts for the reduction of entanglement arising from degenerate states. We present results for several small spin systems. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 16:34:27 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 00:00:49 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Curnoe",
"S. H.",
""
],
[
"Gajera",
"D.",
""
],
[
"Wei",
"C.",
""
]
] |
2506.00095 | Yuchong Li | Yuchong Li, Xiaojun Zeng, Chihua Fang, Jian Yang, Fucang Jia, Lei Zhang | ClinBench-HPB: A Clinical Benchmark for Evaluating LLMs in Hepato-Pancreato-Biliary Diseases | null | null | null | null | cs.CY cs.AI cs.CL | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Hepato-pancreato-biliary (HPB) disorders represent a global public health challenge due to their high morbidity and mortality. Although large language models (LLMs) have shown promising performance in general medical question-answering tasks, the current evaluation benchmarks are mostly derived from standardized examinations or manually designed questions, lacking HPB coverage and clinical cases. To address these issues, we systematically eatablish an HPB disease evaluation benchmark comprising 3,535 closed-ended multiple-choice questions and 337 open-ended real diagnosis cases, which encompasses all the 33 main categories and 465 subcategories of HPB diseases defined in the International Statistical Classification of Diseases, 10th Revision (ICD-10). The multiple-choice questions are curated from public datasets and synthesized data, and the clinical cases are collected from prestigious medical journals, case-sharing platforms, and collaborating hospitals. By evalauting commercial and open-source general and medical LLMs on our established benchmark, namely ClinBench-HBP, we find that while commercial LLMs perform competently on medical exam questions, they exhibit substantial performance degradation on HPB diagnosis tasks, especially on complex, inpatient clinical cases. Those medical LLMs also show limited generalizability to HPB diseases. Our results reveal the critical limitations of current LLMs in the domain of HPB diseases, underscoring the imperative need for future medical LLMs to handle real, complex clinical diagnostics rather than simple medical exam questions. The benchmark will be released at https://clinbench-hpb.github.io. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 11:35:05 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 03:02:20 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 03:25:49 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Li",
"Yuchong",
""
],
[
"Zeng",
"Xiaojun",
""
],
[
"Fang",
"Chihua",
""
],
[
"Yang",
"Jian",
""
],
[
"Jia",
"Fucang",
""
],
[
"Zhang",
"Lei",
""
]
] |
2506.00100 | Ajinkya Kulkarni | Ajinkya Kulkarni and Francisco Teixeira and Enno Hermann and Thomas Rolland and Isabel Trancoso and Mathew Magimai Doss | Children's Voice Privacy: First Steps And Emerging Challenges | Accepted at Interspeech 2025, Netherlands | null | null | null | cs.CY cs.AI cs.CL cs.CR | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Children are one of the most under-represented groups in speech technologies, as well as one of the most vulnerable in terms of privacy. Despite this, anonymization techniques targeting this population have received little attention. In this study, we seek to bridge this gap, and establish a baseline for the use of voice anonymization techniques designed for adult speech when applied to children's voices. Such an evaluation is essential, as children's speech presents a distinct set of challenges when compared to that of adults. This study comprises three children's datasets, six anonymization methods, and objective and subjective utility metrics for evaluation. Our results show that existing systems for adults are still able to protect children's voice privacy, but suffer from much higher utility degradation. In addition, our subjective study displays the challenges of automatic evaluation methods for speech quality in children's speech, highlighting the need for further research. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 13:21:18 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 13:30:41 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Kulkarni",
"Ajinkya",
""
],
[
"Teixeira",
"Francisco",
""
],
[
"Hermann",
"Enno",
""
],
[
"Rolland",
"Thomas",
""
],
[
"Trancoso",
"Isabel",
""
],
[
"Doss",
"Mathew Magimai",
""
]
] |
2506.00140 | Jesse Thibodeau | Jesse Thibodeau, Hadi Nekoei, Afaf Ta\"ik, Janarthanan Rajendran, Golnoosh Farnadi | Balancing Profit and Fairness in Risk-Based Pricing Markets | null | null | null | null | cs.AI cs.LG econ.GN q-fin.EC | http://creativecommons.org/licenses/by/4.0/ | Dynamic, risk-based pricing can systematically exclude vulnerable consumer groups from essential resources such as health insurance and consumer credit. We show that a regulator can realign private incentives with social objectives through a learned, interpretable tax schedule. First, we provide a formal proposition that bounding each firm's \emph{local} demographic gap implicitly bounds the \emph{global} opt-out disparity, motivating firm-level penalties. Building on this insight we introduce \texttt{MarketSim} -- an open-source, scalable simulator of heterogeneous consumers and profit-maximizing firms -- and train a reinforcement learning (RL) social planner (SP) that selects a bracketed fairness-tax while remaining close to a simple linear prior via an $\mathcal{L}_1$ regularizer. The learned policy is thus both transparent and easily interpretable. In two empirically calibrated markets, i.e., U.S. health-insurance and consumer-credit, our planner simultaneously raises demand-fairness by up to $16\%$ relative to unregulated Free Market while outperforming a fixed linear schedule in terms of social welfare without explicit coordination. These results illustrate how AI-assisted regulation can convert a competitive social dilemma into a win-win equilibrium, providing a principled and practical framework for fairness-aware market oversight. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 18:24:08 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 16:06:36 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Thibodeau",
"Jesse",
""
],
[
"Nekoei",
"Hadi",
""
],
[
"Taïk",
"Afaf",
""
],
[
"Rajendran",
"Janarthanan",
""
],
[
"Farnadi",
"Golnoosh",
""
]
] |
2506.00202 | Matthew Kam | Matthew Kam, Cody Miller, Miaoxin Wang, Abey Tidwell, Irene A. Lee, Joyce Malyn-Smith, Beatriz Perez, Vikram Tiwari, Joshua Kenitzer, Andrew Macvean and Erin Barrar | What do professional software developers need to know to succeed in an age of Artificial Intelligence? | 12 pages, 4 figures, software engineering education track of the 2025 ACM international conference on the foundations of software engineering, includes supplementary material i.e. full 50-page occupational profile of the AI-enhanced software developer | null | 10.1145/3696630.3727251 | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Generative AI is showing early evidence of productivity gains for software developers, but concerns persist regarding workforce disruption and deskilling. We describe our research with 21 developers at the cutting edge of using AI, summarizing 12 of their work goals we uncovered, together with 75 associated tasks and the skills & knowledge for each, illustrating how developers use AI at work. From all of these, we distilled our findings in the form of 5 insights. We found that the skills & knowledge to be a successful AI-enhanced developer are organized into four domains (using Generative AI effectively, core software engineering, adjacent engineering, and adjacent non-engineering) deployed at critical junctures throughout a 6-step task workflow. In order to "future proof" developers for this age of AI, on-the-job learning initiatives and computer science degree programs will need to target both "soft" skills and the technical skills & knowledge in all four domains to reskill, upskill and safeguard against deskilling. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 20:14:03 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 17:32:53 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Kam",
"Matthew",
""
],
[
"Miller",
"Cody",
""
],
[
"Wang",
"Miaoxin",
""
],
[
"Tidwell",
"Abey",
""
],
[
"Lee",
"Irene A.",
""
],
[
"Malyn-Smith",
"Joyce",
""
],
[
"Perez",
"Beatriz",
""
],
[
"Tiwari",
"Vikram",
""
],
[
"Kenitzer",
"Joshua",
""
],
[
"Macvean",
"Andrew",
""
],
[
"Barrar",
"Erin",
""
]
] |
2506.00264 | Mohammadamin Shafiei | Mohammadamin Shafiei, Hamidreza Saffari, Nafise Sadat Moosavi | MultiHoax: A Dataset of Multi-hop False-Premise Questions | accepted at ACL Findings 2025 | null | null | null | cs.CL | http://creativecommons.org/licenses/by-sa/4.0/ | As Large Language Models are increasingly deployed in high-stakes domains, their ability to detect false assumptions and reason critically is crucial for ensuring reliable outputs. False-premise questions (FPQs) serve as an important evaluation method by exposing cases where flawed assumptions lead to incorrect responses. While existing benchmarks focus on single-hop FPQs, real-world reasoning often requires multi-hop inference, where models must verify consistency across multiple reasoning steps rather than relying on surface-level cues. To address this gap, we introduce MultiHoax, a benchmark for evaluating LLMs' ability to handle false premises in complex, multi-step reasoning tasks. Our dataset spans seven countries and ten diverse knowledge categories, using Wikipedia as the primary knowledge source to enable factual reasoning across regions. Experiments reveal that state-of-the-art LLMs struggle to detect false premises across different countries, knowledge categories, and multi-hop reasoning types, highlighting the need for improved false premise detection and more robust multi-hop reasoning capabilities in LLMs. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 21:55:18 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 13:23:25 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Shafiei",
"Mohammadamin",
""
],
[
"Saffari",
"Hamidreza",
""
],
[
"Moosavi",
"Nafise Sadat",
""
]
] |
2506.00316 | Atul Ganju | Atul Ganju, Shashaank Aiyer, Ved Sriraman, Karthik Sridharan | Active Learning via Regression Beyond Realizability | null | null | null | null | cs.LG math.ST stat.ML stat.TH | http://creativecommons.org/licenses/by/4.0/ | We present a new active learning framework for multiclass classification based on surrogate risk minimization that operates beyond the standard realizability assumption. Existing surrogate-based active learning algorithms crucially rely on realizability$\unicode{x2014}$the assumption that the optimal surrogate predictor lies within the model class$\unicode{x2014}$limiting their applicability in practical, misspecified settings. In this work we show that under conditions significantly weaker than realizability, as long as the class of models considered is convex, one can still obtain a label and sample complexity comparable to prior work. Despite achieving similar rates, the algorithmic approaches from prior works can be shown to fail in non-realizable settings where our assumption is satisfied. Our epoch-based active learning algorithm departs from prior methods by fitting a model from the full class to the queried data in each epoch and returning an improper classifier obtained by aggregating these models. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 00:04:07 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Ganju",
"Atul",
""
],
[
"Aiyer",
"Shashaank",
""
],
[
"Sriraman",
"Ved",
""
],
[
"Sridharan",
"Karthik",
""
]
] |
2506.00335 | Jingyang He | Jingyang He, Shuai Wang, Ang Li | Recover Experimental Data with Selection Bias using Counterfactual Logic | null | null | null | null | stat.ME cs.AI | http://creativecommons.org/licenses/by/4.0/ | Selection bias, arising from the systematic inclusion or exclusion of certain samples, poses a significant challenge to the validity of causal inference. While Bareinboim et al. introduced methods for recovering unbiased observational and interventional distributions from biased data using partial external information, the complexity of the backdoor adjustment and the method's strong reliance on observational data limit its applicability in many practical settings. In this paper, we formally discover the recoverability of $P(Y^*_{x^*})$ under selection bias with experimental data. By explicitly constructing counterfactual worlds via Structural Causal Models (SCMs), we analyze how selection mechanisms in the observational world propagate to the counterfactual domain. We derive a complete set of graphical and theoretical criteria to determine that the experimental distribution remain unaffected by selection bias. Furthermore, we propose principled methods for leveraging partially unbiased observational data to recover $P(Y^*_{x^*})$ from biased experimental datasets. Simulation studies replicating realistic research scenarios demonstrate the practical utility of our approach, offering concrete guidance for mitigating selection bias in applied causal inference. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 01:23:39 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 17:00:31 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"He",
"Jingyang",
""
],
[
"Wang",
"Shuai",
""
],
[
"Li",
"Ang",
""
]
] |
2506.00353 | Akira Yamamoto | Gerald Dugan, Andrew J. Lankford, Benno List, Shinichiro Michizono, Tatsuya Nakada, Marc Ross, Hiroshi R. Sakai, Steinar Stapnes, Nobuhiro Terunuma, Nicholas Walker, and Akira Yamamoto | ILC250 Cost Update -- 2024 | 15 pages, 7 figures, Backup document for "Status for the International Linear Collider" submitted to European Strategy for Particle Physics Update 2026 | null | null | IDT-EB2025-001-Supplement | physics.acc-ph hep-ex | http://creativecommons.org/licenses/by-nc-sa/4.0/ | The International Linear Collider was conceived as a global project for an energy-frontier electron-positron collider.It employs superconducting RF and nano-beam technologies with a center-of-mass energy of 500 GeV. Its cost was estimated in 2013, based on the Technical Design Report published in 2013.Japan's high-energy community proposed to host the ILC in Japan as a Higgs boson factory at 250 GeV in its first phase, and a revised cost estimate was conducted in 2017 to host it in Japan. However, due to global price increases and currency fluctuations that emerged afterward, the 2017 estimate is now outdated. A new cost evaluation has therefore been performed, according for global inflation tends, exchange rate shifts, and recent experiences in SRF based accelerators. This report describes the cost update performed in 2024. The cost update is included in the ILC Status Report in May 2025, contributing to the ongoing 2026 update of the European Strategy for Particle Physics. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 02:33:16 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 05:00:34 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Dugan",
"Gerald",
""
],
[
"Lankford",
"Andrew J.",
""
],
[
"List",
"Benno",
""
],
[
"Michizono",
"Shinichiro",
""
],
[
"Nakada",
"Tatsuya",
""
],
[
"Ross",
"Marc",
""
],
[
"Sakai",
"Hiroshi R.",
""
],
[
"Stapnes",
"Steinar",
""
],
[
"Terunuma",
"Nobuhiro",
""
],
[
"Walker",
"Nicholas",
""
],
[
"Yamamoto",
"Akira",
""
]
] |
2506.00486 | Jun Wu | Jun Wu, Yirong Xiong, Jiangtao Wen, Yuxing Han | It Takes a Good Model to Train a Good Model: Generalized Gaussian Priors for Optimized LLMs | null | null | null | null | cs.LG cs.AI stat.ML | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Despite rapid advancements in the research and deployment of large language models (LLMs), the statistical distribution of model parameters, as well as their influence on initialization, training dynamics, and downstream efficiency, has received surprisingly little attention. A recent work introduced BackSlash, a training-time compression algorithm. It first demonstrated that pre-trained LLM parameters follow generalized Gaussian distributions (GGDs) better. By optimizing GG priors during training, BackSlash can reduce parameters by up to 90\% with minimal performance loss. Building on this foundational insight, we propose a unified, end-to-end framework for LLM optimization based on the GG model. Our contributions are threefold: (1) GG-based initialization scheme that aligns with the statistical structure of trained models, resulting in faster convergence and improved accuracy; (2) DeepShape, a post-training regularization method that reshapes weight distributions to match a GG profile, improving compressibility with minimized degradation in performance; and (3) RF8, a compact and hardware-efficient 8-bit floating-point format designed for GG-distributed-initialized BackSlash training, enabling low-cost inference without compromising accuracy. Experiments across diverse model architectures show that our framework consistently yields smaller and faster models that match or outperform standard training baselines. By grounding LLM development in principled statistical modeling, this work forges a new path toward efficient, scalable, and hardware-aware AI systems. The code is available on our project page: https://huggingface.co/spaces/shifeng3711/gg_prior. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 09:49:17 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 06:38:16 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 08:00:08 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Wu",
"Jun",
""
],
[
"Xiong",
"Yirong",
""
],
[
"Wen",
"Jiangtao",
""
],
[
"Han",
"Yuxing",
""
]
] |
2506.00493 | Tim Bedding | Hans Kjeldsen, Timothy R. Bedding, Yaguang Li, Frank Grundahl, Mads Fredslund Andersen, Duncan J. Wright, Jack Soutter, Robert Wittenmyer, Claudia Reyes, Dennis Stello, Courtney Crawford, Yixiao Zhou, Mathieu Clerte, Pere L. Palle, Sergio Simon-Diaz, Joergen Christensen-Dalsgaard, Rasmus Handberg, Hasse Hansen, Paul Heeren, Jens Jessen-Hansen, Mikkel Lund, Mia S. Lundkvist, Karsten Brogaard, Rene Tronsgaard, Jonatan Rudrasingam, Luca Casagrande, Jonathan Horner, Daniel Huber, John Lattanzio, Sarah L. Martell, Simon J. Murphy | Asteroseismology of the G8 subgiant beta Aquilae with SONG-Tenerife, SONG-Australia and TESS | submitted to A&A; changes made in response to referee's report | null | null | null | astro-ph.SR | http://creativecommons.org/licenses/by/4.0/ | We present radial velocities of the G8 subgiant star beta Aql obtained in 2022 and 2023 using SONG-Tenerife and SONG-Australia. We also analysed a sector of TESS photometry that overlapped with the 2022 SONG data. We processed the time series to assign weights and to remove bad data points. The power spectrum clearly shows solar-like oscillations centred at 430 muHz. The TESS data show the oscillations at lower signal-to-noise because photometric measurements are much more affected by the granulation background than are radial velocities. The simultaneous observations in velocity and photometry represent the best such measurements for any star apart from the Sun. They allowed us to measure the ratio between the two to be 26.6+/-3.1 ppm/(m/s). We measured this ratio for the Sun from published SOHO data to be 19.5+/-0.7 ppm/(m/s) and, after accounting for the difference in effective temperatures of the beta Aql and the Sun, these values align with expectations. In both the Sun and beta Aql, the photometry-to-velocity ratio appears to be a function of frequency. We also measured the phase shift of the oscillations in beta Aql between SONG velocities and TESS photometry to be -113+/-7 deg, which agrees within uncertainties with the value for the Sun and also with a 3-D simulation of a star with similar properties to beta Aql. Importantly for exoplanet searches, we argue that simultaneous photometry can be used to predict the contribution of oscillations to radial velocities. We identified and measured frequencies for 22 oscillation modes in beta Aql with angular degrees of l=0, 1 and 2. Asteroseismic modelling yielded an excellent fit to the frequencies, giving accurate values for mass and age, placing quite strong constraints on the mixing-length parameter. Finally, the oscillation properties of beta Aql are very similar to subgiants and low-luminosity RGB stars in the open cluster M67. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 10:01:21 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 23:17:31 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Kjeldsen",
"Hans",
""
],
[
"Bedding",
"Timothy R.",
""
],
[
"Li",
"Yaguang",
""
],
[
"Grundahl",
"Frank",
""
],
[
"Andersen",
"Mads Fredslund",
""
],
[
"Wright",
"Duncan J.",
""
],
[
"Soutter",
"Jack",
""
],
[
"Wittenmyer",
"Robert",
""
],
[
"Reyes",
"Claudia",
""
],
[
"Stello",
"Dennis",
""
],
[
"Crawford",
"Courtney",
""
],
[
"Zhou",
"Yixiao",
""
],
[
"Clerte",
"Mathieu",
""
],
[
"Palle",
"Pere L.",
""
],
[
"Simon-Diaz",
"Sergio",
""
],
[
"Christensen-Dalsgaard",
"Joergen",
""
],
[
"Handberg",
"Rasmus",
""
],
[
"Hansen",
"Hasse",
""
],
[
"Heeren",
"Paul",
""
],
[
"Jessen-Hansen",
"Jens",
""
],
[
"Lund",
"Mikkel",
""
],
[
"Lundkvist",
"Mia S.",
""
],
[
"Brogaard",
"Karsten",
""
],
[
"Tronsgaard",
"Rene",
""
],
[
"Rudrasingam",
"Jonatan",
""
],
[
"Casagrande",
"Luca",
""
],
[
"Horner",
"Jonathan",
""
],
[
"Huber",
"Daniel",
""
],
[
"Lattanzio",
"John",
""
],
[
"Martell",
"Sarah L.",
""
],
[
"Murphy",
"Simon J.",
""
]
] |
2506.00521 | Shida Wang | Shida Wang, Jalal Fadili, Peter Ochs | Convergence rates of regularized quasi-Newton methods without strong convexity | arXiv admin note: text overlap with arXiv:2410.11676 | null | null | null | math.OC | http://creativecommons.org/licenses/by/4.0/ | In this paper, we study convergence rates of the cubic regularized proximal quasi-Newton method (\csr) for solving non-smooth additive composite problems that satisfy the so-called Kurdyka-\L ojasiewicz (K\L ) property with respect to some desingularization function $\phi$ rather than strong convexity. After a number of iterations $k_0$, Cubic SR1 PQN exhibits non-asymptotic explicit super-linear convergence rates for any $k\geq k_0$. In particular, when $\phi(t)=ct^{1/2}$, Cubic SR1 PQN has a convergence rate of order $\left(\frac{C}{(k-k_0)^{1/2}}\right)^{(k-k_0)/2}$, where $k$ is the number of iterations and $C>0$ is a constant. For the special case, i.e. functions which satisfy \L ojasiewicz inequality, the rate becomes global and non-asymptotic. This work presents, for the first time, non-asymptotic explicit convergence rates of regularized (proximal) SR1 quasi-Newton methods applied to non-convex non-smooth problems with K\L\ property. Actually, the rates are novel even in the smooth non-convex case. Notably, we achieve this without employing line search or trust region strategies, without assuming the Dennis-Mor\'e condition, without any assumptions on quasi-Newton metrics and without assuming strong convexity. Furthermore, for convex problems, we focus on a more tractable gradient regularized quasi-Newton method (Grad SR1 PQN) which can achieve results similar to those obtained with cubic regularization. We also demonstrate, for the first time, the non-asymptotic super-linear convergence rate of Grad SR1 PQN for solving convex problems with the help of the \L ojasiewicz inequality instead of strong convexity. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 11:48:12 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 13:35:21 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Wang",
"Shida",
""
],
[
"Fadili",
"Jalal",
""
],
[
"Ochs",
"Peter",
""
]
] |
2506.00524 | Hui Li | Hui Li, Jie Xie, Hyukjoon Kwon, Yixin Zhao, M.S. Kim, Lijian Zhang | Experimental demonstration of generalized quantum fluctuation theorems in the presence of coherence | null | Sci. Adv. 11, eadq6014(2025) | 10.1126/sciadv.adq6014 | null | quant-ph | http://creativecommons.org/licenses/by/4.0/ | Fluctuation theorems have elevated the second law of thermodynamics to a statistical realm by establishing a connection between time-forward and time-reversal probabilities, providing invaluable insight into nonequilibrium dynamics. While well established in classical systems, their quantum generalization, incorporating coherence and the diversity of quantum noise, remains open. We report the experimental validation of a quantum fluctuation theorem (QFT) in a photonic system, applicable to general quantum processes with nonclassical characteristics, including quasi-probabilistic descriptions of entropy production and multiple time-reversal processes. Our experiment confirms that the ratio between the quasi-probabilities of the time-forward and any multiple time-reversal processes obeys a generalized Crooks QFT. Moreover, coherence induced by a quantum process leads to the imaginary components of quantum entropy production, governing the phase factor in the QFT. These findings underscore the fundamental symmetry between a general quantum process and its time reversal, providing an elementary toolkit to explore noisy quantum information processing. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 12:00:59 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Li",
"Hui",
""
],
[
"Xie",
"Jie",
""
],
[
"Kwon",
"Hyukjoon",
""
],
[
"Zhao",
"Yixin",
""
],
[
"Kim",
"M. S.",
""
],
[
"Zhang",
"Lijian",
""
]
] |
2506.00539 | Ruihan Yang | Ruihan Yang, Yikai Zhang, Aili Chen, Xintao Wang, Siyu Yuan, Jiangjie Chen, Deqing Yang, Yanghua Xiao | ARIA: Training Language Agents with Intention-Driven Reward Aggregation | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) have enabled agents to perform complex reasoning and decision-making through free-form language interactions. However, in open-ended language action environments (e.g., negotiation or question-asking games), the action space can be formulated as a joint distribution over tokens, resulting in an exponentially large action space. Sampling actions in such a space can lead to extreme reward sparsity, which brings large reward variance, hindering effective reinforcement learning (RL). To address this, we propose ARIA, a method that Aggregates Rewards in Intention space to enable efficient and effective language Agents training. ARIA aims to project natural language actions from the high-dimensional joint token distribution space into a low-dimensional intention space, where semantically similar actions are clustered and assigned shared rewards. This intention-aware reward aggregation reduces reward variance by densifying reward signals, fostering better policy optimization. Extensive experiments demonstrate that ARIA not only significantly reduces policy gradient variance, but also delivers substantial performance gains of an average of 9.95% across four downstream tasks, consistently outperforming offline and online RL baselines. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 12:54:49 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 13:39:54 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Yang",
"Ruihan",
""
],
[
"Zhang",
"Yikai",
""
],
[
"Chen",
"Aili",
""
],
[
"Wang",
"Xintao",
""
],
[
"Yuan",
"Siyu",
""
],
[
"Chen",
"Jiangjie",
""
],
[
"Yang",
"Deqing",
""
],
[
"Xiao",
"Yanghua",
""
]
] |
2506.00605 | Ruiming Min | Ruiming Min, Minghao Liu | ABCDEFGH: An Adaptation-Based Convolutional Neural Network-CycleGAN Disease-Courses Evolution Framework Using Generative Models in Health Education | All authors did not agree to submitting this work. This version of the report contains misinformation and is not ready to share | null | null | null | eess.IV cs.CV | http://creativecommons.org/licenses/by/4.0/ | With the advancement of modern medicine and the development of technologies such as MRI, CT, and cellular analysis, it has become increasingly critical for clinicians to accurately interpret various diagnostic images. However, modern medical education often faces challenges due to limited access to high-quality teaching materials, stemming from privacy concerns and a shortage of educational resources (Balogh et al., 2015). In this context, image data generated by machine learning models, particularly generative models, presents a promising solution. These models can create diverse and comparable imaging datasets without compromising patient privacy, thereby supporting modern medical education. In this study, we explore the use of convolutional neural networks (CNNs) and CycleGAN (Zhu et al., 2017) for generating synthetic medical images. The source code is available at https://github.com/mliuby/COMP4211-Project. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 15:32:58 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 17:27:24 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Min",
"Ruiming",
""
],
[
"Liu",
"Minghao",
""
]
] |
2506.00618 | Jinyi Yang | Jingyi Yang, Shuai Shao, Dongrui Liu, Jing Shao | RiOSWorld: Benchmarking the Risk of Multimodal Computer-Use Agents | 40 pages, 6 figures, Project Page: https://yjyddq.github.io/RiOSWorld.github.io/ | null | null | null | cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | With the rapid development of multimodal large language models (MLLMs), they are increasingly deployed as autonomous computer-use agents capable of accomplishing complex computer tasks. However, a pressing issue arises: Can the safety risk principles designed and aligned for general MLLMs in dialogue scenarios be effectively transferred to real-world computer-use scenarios? Existing research on evaluating the safety risks of MLLM-based computer-use agents suffers from several limitations: it either lacks realistic interactive environments, or narrowly focuses on one or a few specific risk types. These limitations ignore the complexity, variability, and diversity of real-world environments, thereby restricting comprehensive risk evaluation for computer-use agents. To this end, we introduce \textbf{RiOSWorld}, a benchmark designed to evaluate the potential risks of MLLM-based agents during real-world computer manipulations. Our benchmark includes 492 risky tasks spanning various computer applications, involving web, social media, multimedia, os, email, and office software. We categorize these risks into two major classes based on their risk source: (i) User-originated risks and (ii) Environmental risks. For the evaluation, we evaluate safety risks from two perspectives: (i) Risk goal intention and (ii) Risk goal completion. Extensive experiments with multimodal agents on \textbf{RiOSWorld} demonstrate that current computer-use agents confront significant safety risks in real-world scenarios. Our findings highlight the necessity and urgency of safety alignment for computer-use agents in real-world computer manipulation, providing valuable insights for developing trustworthy computer-use agents. Our benchmark is publicly available at https://yjyddq.github.io/RiOSWorld.github.io/. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 16:04:59 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 08:23:57 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Yang",
"Jingyi",
""
],
[
"Shao",
"Shuai",
""
],
[
"Liu",
"Dongrui",
""
],
[
"Shao",
"Jing",
""
]
] |
2506.00631 | Viktor Zozulia | Viktor D. Zozulia, Natalia Ya. Sotnikova, and Anton A. Smirnov | Phase-space distortion as a key to unraveling galactic bar buckling | 7 pages, 4 figures, Accepted in the A&A Letters | null | 10.1051/0004-6361/202554837 | null | astro-ph.GA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | For the first time, we investigate the resonant structure of $N$-body galactic bar at the stage of buckling using action-angle variables. We studied the evolution of vertical actions ($J_z$) and angles associated with vertical resonance ($\theta_\mathrm{res}=\theta_z - \theta_R$) for all orbits in the bar. For this purpose, we divide the orbits into types according to the behavior (libration or circulation) of their resonant angle with respect to fixed points $\theta_\mathrm{res}=0$ and $\pi$ (vertical resonance). We show that during buckling, flat bar orbits circulating with increasing $\theta_\mathrm{res}$ transformed into banana-shaped librating orbits (resonant capture) or circulating orbits with decreasing $\theta_\mathrm{res}$ (resonant heating). The orbital transformation is accompanied by an increase in $J_z$ and the formation of a boxy/peanut-shaped (B/PS) bulge. During buckling, the phase space $J_z - \theta_\mathrm{res}$ undergoes a distortion creating an asymmetry in the position of the fixed points $\theta_\mathrm{res}=0$ and $\pi$ and in banana-shaped orbits near these points. The fixed point $\theta_\mathrm{res}=0$ may disappear completely. This also breaks the symmetry between the orbits, which are captured into resonance or go into circulation with decreasing $\theta_\mathrm{res}$ near $\theta_\mathrm{res}=0$ and $\pi$. At the same time, near $\theta_\mathrm{res}=0$, banana-shaped orbits with low vertical action $J_z$ appear. This reopens the path of orbital transformation through the fixed point $\theta_\mathrm{res}=0$. The phase space transformation and orbit transformation occur in a coordinated manner and lead to smoothing of phase space perturbations and restoration of symmetry between orbits. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 16:39:11 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 07:25:29 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Zozulia",
"Viktor D.",
""
],
[
"Sotnikova",
"Natalia Ya.",
""
],
[
"Smirnov",
"Anton A.",
""
]
] |
2506.00691 | Junaid Muzaffar | Junaid Muzaffar, Khubaib Ahmed, Ingo Frommholz, Zeeshan Pervez, Ahsan ul Haq | Optimizing Sensory Neurons: Nonlinear Attention Mechanisms for Accelerated Convergence in Permutation-Invariant Neural Networks for Reinforcement Learning | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Training reinforcement learning (RL) agents often requires significant computational resources and extended training times. To address this, we build upon the foundation laid by Google Brain's Sensory Neuron, which introduced a novel neural architecture for reinforcement learning tasks that maintained permutation in-variance in the sensory neuron system. While the baseline model demonstrated significant performance improvements over traditional approaches, we identified opportunities to enhance the efficiency of the learning process further. We propose a modified attention mechanism incorporating a non-linear transformation of the key vectors (K) using a mapping function, resulting in a new set of key vectors (K'). This non-linear mapping enhances the representational capacity of the attention mechanism, allowing the model to encode more complex feature interactions and accelerating convergence without compromising performance. Our enhanced model demonstrates significant improvements in learning efficiency, showcasing the potential for non-linear attention mechanisms in advancing reinforcement learning algorithms. | [
{
"version": "v1",
"created": "Sat, 31 May 2025 19:50:37 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 15:54:50 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Muzaffar",
"Junaid",
""
],
[
"Ahmed",
"Khubaib",
""
],
[
"Frommholz",
"Ingo",
""
],
[
"Pervez",
"Zeeshan",
""
],
[
"Haq",
"Ahsan ul",
""
]
] |
2506.01016 | Dhruv Sreenivas | Olya Mastikhina, Dhruv Sreenivas, Pablo Samuel Castro | Optimistic critics can empower small actors | RLC 2025 | null | null | null | cs.LG stat.ML | http://creativecommons.org/licenses/by/4.0/ | Actor-critic methods have been central to many of the recent advances in deep reinforcement learning. The most common approach is to use symmetric architectures, whereby both actor and critic have the same network topology and number of parameters. However, recent works have argued for the advantages of asymmetric setups, specifically with the use of smaller actors. We perform broad empirical investigations and analyses to better understand the implications of this and find that, in general, smaller actors result in performance degradation and overfit critics. Our analyses suggest poor data collection, due to value underestimation, as one of the main causes for this behavior, and further highlight the crucial role the critic can play in alleviating this pathology. We explore techniques to mitigate the observed value underestimation, which enables further research in asymmetric actor-critic methods. | [
{
"version": "v1",
"created": "Sun, 01 Jun 2025 14:00:03 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 15:56:22 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Mastikhina",
"Olya",
""
],
[
"Sreenivas",
"Dhruv",
""
],
[
"Castro",
"Pablo Samuel",
""
]
] |
2506.01056 | Xiang Fei | Xiang Fei, Xiawu Zheng, Hao Feng | MCP-Zero: Proactive Toolchain Construction for LLM Agents from Scratch | null | null | null | null | cs.AI cs.SE | http://creativecommons.org/licenses/by/4.0/ | Function-calling has enabled large language models (LLMs) to act as tool-using agents, but injecting thousands of tool schemas into the prompt is costly and error-prone. We introduce MCP-Zero, a proactive agent framework that lets the LLM itself decide when and which external tools to retrieve, thereby assembling a task-specific toolchain from scratch. The framework is built upon three components: (1) Proactive Tool Request, where the model emits a structured $\left<\operatorname{tool\_assistant}\right>$ block that explicitly specifies the desired server and task; (2) Hierarchical Vector Routing, a coarse-to-fine retrieval algorithm that first selects candidate servers and then ranks tools within each server based on the semantic similarity; (3) Iterative Proactive Invocation, enabling multi-round, cross-domain toolchain construction with minimal context overhead, and allowing the model to iteratively revise its request when the returned tools are insufficient. To evaluate our approach we also compile MCP-tools, a retrieval dataset comprising 308 MCP servers and 2,797 tools extracted from the official Model-Context-Protocol repository and normalized into a unified JSON schema. Experiments show that MCP-Zero (i) effectively addresses the context overhead problem of existing methods and accurately selects the correct tool from a pool of nearly 3,000 candidates (248.1k tokens); (ii) reduces token consumption by 98\% on the APIbank while maintaining high accuracy; and (iii) supports multi-turn tool invocation with consistent accuracy across rounds. | [
{
"version": "v1",
"created": "Sun, 01 Jun 2025 15:48:53 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 06:37:09 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Fei",
"Xiang",
""
],
[
"Zheng",
"Xiawu",
""
],
[
"Feng",
"Hao",
""
]
] |
2506.01144 | Itay Hazan | Ariel Shaulov, Itay Hazan, Lior Wolf, Hila Chefer | FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Text-to-video diffusion models are notoriously limited in their ability to model temporal aspects such as motion, physics, and dynamic interactions. Existing approaches address this limitation by retraining the model or introducing external conditioning signals to enforce temporal consistency. In this work, we explore whether a meaningful temporal representation can be extracted directly from the predictions of a pre-trained model without any additional training or auxiliary inputs. We introduce FlowMo, a novel training-free guidance method that enhances motion coherence using only the model's own predictions in each diffusion step. FlowMo first derives an appearance-debiased temporal representation by measuring the distance between latents corresponding to consecutive frames. This highlights the implicit temporal structure predicted by the model. It then estimates motion coherence by measuring the patch-wise variance across the temporal dimension and guides the model to reduce this variance dynamically during sampling. Extensive experiments across multiple text-to-video models demonstrate that FlowMo significantly improves motion coherence without sacrificing visual quality or prompt alignment, offering an effective plug-and-play solution for enhancing the temporal fidelity of pre-trained video diffusion models. | [
{
"version": "v1",
"created": "Sun, 01 Jun 2025 19:55:33 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 07:45:25 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Shaulov",
"Ariel",
""
],
[
"Hazan",
"Itay",
""
],
[
"Wolf",
"Lior",
""
],
[
"Chefer",
"Hila",
""
]
] |
2506.01186 | Jonathan Hillman | Jonathan A. Hillman | 3-manifolds with more than one abelian embedding | v2 rewritten to delete overclaim (re $\beta\leq2$) and give new example with 5 inequivalent abelian embeddings | null | null | null | math.GT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We construct 3-manifolds which have at least two inequivalent embeddings such that both complementary regions have abelian fundamental group. | [
{
"version": "v1",
"created": "Sun, 01 Jun 2025 21:46:17 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 03:38:04 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Hillman",
"Jonathan A.",
""
]
] |
2506.01224 | John Smutny | John W. Smutny | Dirty and Clean-Label attack detection using GAN discriminators | 13 pages total. Appendix starts on page 10 | null | null | null | cs.CV eess.IV | http://creativecommons.org/licenses/by/4.0/ | Gathering enough images to train a deep computer vision model is a constant challenge. Unfortunately, collecting images from unknown sources can leave your model s behavior at risk of being manipulated by a dirty-label or clean-label attack unless the images are properly inspected. Manually inspecting each image-label pair is impractical and common poison-detection methods that involve re-training your model can be time consuming. This research uses GAN discriminators to protect a single class against mislabeled and different levels of modified images. The effect of said perturbation on a basic convolutional neural network classifier is also included for reference. The results suggest that after training on a single class, GAN discriminator s confidence scores can provide a threshold to identify mislabeled images and identify 100% of the tested poison starting at a perturbation epsilon magnitude of 0.20, after decision threshold calibration using in-class samples. Developers can use this report as a basis to train their own discriminators to protect high valued classes in their CV models. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 00:32:07 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 23:21:56 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Smutny",
"John W.",
""
]
] |
2506.01226 | Nicholas Barbara | Nicholas H. Barbara, Ruigang Wang, Alexandre Megretski, Ian R. Manchester | React to Surprises: Stable-by-Design Neural Feedback Control and the Youla-REN | null | null | null | null | eess.SY cs.LG cs.SY | http://creativecommons.org/licenses/by/4.0/ | We study parameterizations of stabilizing nonlinear policies for learning-based control. We propose a structure based on a nonlinear version of the Youla-Kucera parameterization combined with robust neural networks such as the recurrent equilibrium network (REN). The resulting parameterizations are unconstrained, and hence can be searched over with first-order optimization methods, while always ensuring closed-loop stability by construction. We study the combination of (a) nonlinear dynamics, (b) partial observation, and (c) incremental closed-loop stability requirements (contraction and Lipschitzness). We find that with any two of these three difficulties, a contracting and Lipschitz Youla parameter always leads to contracting and Lipschitz closed loops. However, if all three hold, then incremental stability can be lost with exogenous disturbances. Instead, a weaker condition is maintained, which we call d-tube contraction and Lipschitzness. We further obtain converse results showing that the proposed parameterization covers all contracting and Lipschitz closed loops for certain classes of nonlinear systems. Numerical experiments illustrate the utility of our parameterization when learning controllers with built-in stability certificates for: (i) "economic" rewards without stabilizing effects; (ii) short training horizons; and (iii) uncertain systems. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 00:36:24 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 01:56:15 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Barbara",
"Nicholas H.",
""
],
[
"Wang",
"Ruigang",
""
],
[
"Megretski",
"Alexandre",
""
],
[
"Manchester",
"Ian R.",
""
]
] |
2506.01294 | Alexander Ibrahim | Alexander Ibrahim, Lecheng Wang, Tom Halverson, Robert J Le Roy, Pierre-Nicholas Roy | Equation of State and First Principles Prediction of the Vibrational Matrix Shift of Solid Parahydrogen | null | J. Chem. Phys. 151, 244501 (2019) | 10.1063/1.5131329 | null | physics.chem-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We generate the equation of state (EOS) of solid parahydrogen using a path-integral Monte Carlo (PIMC) simulation based on a highly accurate first-principles adiabatic hindered rotor (AHR) potential energy curve for the parahydrogen dimer. The EOS curves for the fcc and hcp structures of solid parahydrogen near the equilibrium density show that the hcp structure is the more stable of the two, in agreement with experiment. To accurately reproduce the structural and energy properties of solid parahydrogen, we eliminated by extrapolation the systematic errors associated with the choice of simulation parameters used in the PIMC calculation. We also investigate the temperature dependence of the EOS curves, and the invariance of the equilibrium density with temperature is satisfyingly reproduced. The pressure as a function of density, and the compressibility as a function of pressure, are both calculated using the obtained EOS and are compared with previous simulation results and experiments. We also report the first ever a priori prediction of a vibrational matrix shift from first-principles two-body potential functions, and its result for the equilibrium state agrees well with experiment. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 04:02:51 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 18:59:36 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Ibrahim",
"Alexander",
""
],
[
"Wang",
"Lecheng",
""
],
[
"Halverson",
"Tom",
""
],
[
"Roy",
"Robert J Le",
""
],
[
"Roy",
"Pierre-Nicholas",
""
]
] |
2506.01297 | Yulun (Chris) Zhou | Ya Wen, Jixuan Cai, Qiyao Ma, Linyan Li, Xinhua Chen, Chris Webster, Yulun Zhou | MobCLIP: Learning General-purpose Geospatial Representation at Scale | null | null | null | null | cs.AI | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Representation learning of geospatial locations remains a core challenge in achieving general geospatial intelligence. Current embedding methods often lack versatility, limiting their utility across diverse tasks in both human and natural domains. We present MobCLIP, the first nationwide general-purpose location encoder, integrating an unprecedented diversity of data modalities through effective and scalable multimodal fusion. Adopting a novel CLIP-based architecture, our framework aligns 100M+ POIs, nationwide remote sensing imagery, and structured demographic statistics with a billion-edge mobility graph. By tokenizing spatial locations into grid cells inspired by Vision Transformers, we establish a unified representation space bridging mobility patterns and multimodal features. To rigorously evaluate the general-purpose effectiveness of MobCLIP, we construct a benchmark dataset composed of 11 downstream prediction tasks across social, economic, and natural domains. Experiments show that MobCLIP, with four input modalities and a compact 128-dimensional representation space, achieves significantly superior general-purpose predictive performances than state-of-the-art models by an average of 35%. Thanks to the effective integration of human-centric modalities, the performance gain is particularly profound in human-centric tasks, such as energy consumption (+260%), offline retail consumption amount (+98%), and crime cases (+95%) predictions. Echoing LLM scaling laws, we further demonstrate the scaling behavior in geospatial representation learning. We open-source code and pretrained models at: https://github.com/ylzhouchris/MobCLIP. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 04:14:03 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 16:53:41 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 02:07:51 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Wen",
"Ya",
""
],
[
"Cai",
"Jixuan",
""
],
[
"Ma",
"Qiyao",
""
],
[
"Li",
"Linyan",
""
],
[
"Chen",
"Xinhua",
""
],
[
"Webster",
"Chris",
""
],
[
"Zhou",
"Yulun",
""
]
] |
2506.01415 | Hiroaki Kanno | Hiroaki Kanno, Ryo Ohkawa and Jun'ichi Shiraishi | Super Macdonald polynomials and BPS state counting on the blow-up | 55 pages | null | null | null | hep-th math-ph math.MP math.QA math.RT | http://creativecommons.org/licenses/by/4.0/ | We explore the relation of the super Macdonald polynomials and the BPS state counting on the blow-up of $\mathbb{P}^2$, which is mathematically described by framed stable perverse coherent sheaves. Fixed points of the torus action on the moduli space of BPS states are labeled by super partitions. From the equivariant character of the tangent space at the fixed points we can define the Nekrasov factor for a pair of super partitions, which is used for the localization computation of the partition function. The Nekrasov factor also allows us to compute matrix elements of the action of the quantum toroidal algebra of type $\mathfrak{gl}_{1|1}$ on the $K$ group of the moduli space. We confirm that these matrix elements are consistent with the Pieri rule of the super Macdonald polynomial. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 08:14:28 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Kanno",
"Hiroaki",
""
],
[
"Ohkawa",
"Ryo",
""
],
[
"Shiraishi",
"Jun'ichi",
""
]
] |
2506.01677 | Guillermo Garc\'ia-S\'aez | Jos\'e C. Bellido and Javier Cueto and Guillermo Garc\'ia-S\'aez | Compact embeddings of Bessel Potential Spaces | null | null | null | null | math.FA math.AP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Bessel potential spaces have gained renewed interest due to their robust structural properties and applications in fractional partial differential equations (PDEs). These spaces, derived through complex interpolation between Lebesgue and Sobolev spaces, are closely related to the Riesz fractional gradient. Recent studies have demonstrated continuous and compact embeddings of Bessel potential spaces into Lebesgue spaces. This paper extends these findings by addressing the compactness of continuous embeddings from the perspective of abstract interpolation theory. We present three distinct proofs, leveraging compactness results, translation estimates, and the relationship between Gagliardo and Bessel spaces. Our results provide a deeper understanding of the functional analytic properties of Bessel potential spaces and their applications in fractional PDEs. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 13:47:00 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 10:33:48 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Bellido",
"José C.",
""
],
[
"Cueto",
"Javier",
""
],
[
"García-Sáez",
"Guillermo",
""
]
] |
2506.01714 | Giulia Forghieri Dr. | Jing Li, Giulia Forghieri, Danny Geelen, Patrick du Jardin, Patrick H. Brown | The optimization of crop response to climatic stress through modulation of plant stress response mechanisms. Opportunities for biostimulants and plant hormones to meet climate challenges | null | Nature Plants 2025 | null | null | q-bio.BM q-bio.GN q-bio.MN | http://creativecommons.org/licenses/by/4.0/ | Climate change is a major threat to crop potential and is characterized by both long-term shifts in temperature and precipitation patterns as well as increased occurrence of extreme weather events, these extreme weather events are the most immediate and intractable threat to agriculture. Crop resilience in the face of stress depends upon the speed and effectiveness with which plants and cropping systems sense and respond to that stress. A variety of agronomic practices including breeding, exogenous inputs (nutrients, water, biostimulants and others) and shifts in cultivation practice have been used to influence plant stress response to achieve the goal of increased plant and cropping system resilience. Traditional breeding is a powerful tool that has resulted in stable and long-term cultivar improvements but is often too slow and complex to meet the diverse, complex and unpredictable challenges of climate induced stresses. Increased inputs (water, nutrients, pesticides etc.) and management strategies (cropping system choice, soil management etc.) can alleviate stress but are often constrained by cost and availability of inputs. Exogenous biostimulants, microbials and plant hormones have shown great promise as mechanisms to optimize natural plant resilience resulting in immediate but non-permanent improvements in plant responses to climate induced stresses. The failure to modernize regulatory frameworks for the use of biostimulants in agriculture will constrain the development of safe effective tools and deprive growers of means to respond to the vagaries of climate change. Here we discuss the scientific rationale for eliminating the regulatory barriers that constrain the potential for biostimulants or products that modulate plant regulatory networks to address climate change challenges and propose a framework for enabling legislation to strengthen cropping system resilience. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 14:22:14 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Li",
"Jing",
""
],
[
"Forghieri",
"Giulia",
""
],
[
"Geelen",
"Danny",
""
],
[
"Jardin",
"Patrick du",
""
],
[
"Brown",
"Patrick H.",
""
]
] |
2506.01723 | Soyoung Oh | Soyoung Oh, Xinting Huang, Mathis Pink, Michael Hahn, Vera Demberg | Tug-of-war between idiom's figurative and literal meanings in LLMs | null | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by/4.0/ | Idioms present a unique challenge for language models due to their non-compositional figurative meanings, which often strongly diverge from the idiom's literal interpretation. This duality requires a model to learn representing and deciding between the two meanings to interpret an idiom in a figurative sense, or literally. In this paper, we employ tools from mechanistic interpretability to trace how a large pretrained causal transformer (LLama3.2-1B-base) deals with this ambiguity. We localize three steps of idiom processing: First, the idiom's figurative meaning is retrieved in early attention and MLP sublayers. We identify specific attention heads which boost the figurative meaning of the idiom while suppressing the idiom's literal interpretation. The model subsequently represents the figurative representation through an intermediate path. Meanwhile, a parallel bypass route forwards literal interpretation, ensuring that a both reading remain available. Overall, our findings provide a mechanistic evidence for idiom comprehension in an autoregressive transformer. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 14:29:46 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 12:05:01 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Oh",
"Soyoung",
""
],
[
"Huang",
"Xinting",
""
],
[
"Pink",
"Mathis",
""
],
[
"Hahn",
"Michael",
""
],
[
"Demberg",
"Vera",
""
]
] |
2506.01832 | Chin Ho Lee | Chin Ho Lee, Emanuele Viola | Pseudorandom bits for non-commutative programs | Minor fixes | null | null | null | cs.CC | http://creativecommons.org/licenses/by/4.0/ | We obtain new explicit pseudorandom generators for several computational models involving groups. Our main results are as follows:
1. We consider read-once group-products over a finite group $G$, i.e., tests of the form $\prod_{i=1}^n g_i^{x_i}$ where $g_i\in G$, a special case of read-once permutation branching programs. We give generators with optimal seed length $c_G \log(n/\varepsilon)$ over any $p$-group. The proof uses the small-bias plus noise paradigm, but derandomizes the noise to avoid the recursion in previous work. Our generator works when the bits are read in any order. Previously for any non-commutative group the best seed length was $\ge\log n\log(1/\varepsilon)$, even for a fixed order.
2. We give a reduction that "lifts" suitable generators for group products over $G$ to a generator that fools width-$w$ block products, i.e., tests of the form $\prod g_i^{f_i}$ where the $f_i$ are arbitrary functions on disjoint blocks of $w$ bits. Block products generalize several previously studied classes. The reduction applies to groups that are mixing in a representation-theoretic sense that we identify.
3. Combining (2) with (1) and other works we obtain new generators for block products over the quaternions or over any commutative group, with nearly optimal seed length. In particular, we obtain generators for read-once polynomials modulo any fixed $m$ with nearly optimal seed length. Previously this was known only for $m=2$.
4. We give a new generator for products over "mixing groups." The construction departs from previous work and uses representation theory. For constant error, we obtain optimal seed length, improving on previous work (which applied to any group).
This paper identifies a challenge in the area that is reminiscent of a roadblock in circuit complexity -- handling composite moduli -- and points to several classes of groups to be attacked next. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 16:21:30 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 13:44:02 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Lee",
"Chin Ho",
""
],
[
"Viola",
"Emanuele",
""
]
] |
2506.01837 | Moosung Lee | Moosung Lee, Benjamin A. Stickler, Thomas Pertsch, Sungkun Hong | Inverse Microparticle Design for Enhanced Optical Trapping and Detection Efficiency in All Six Degrees of Freedom | null | null | null | null | physics.optics quant-ph | http://creativecommons.org/licenses/by/4.0/ | Achieving quantum-limited motional control of optically trapped particles beyond the sub-micrometer scale is an outstanding problem in levitated optomechanics. A key obstacle is solving the light scattering problem and identifying particle geometries that allow stable trapping and efficient motional detection of their center of mass and rotational motion in three dimensions. Here, we present a computational framework that combines an efficient electromagnetic scattering solver with the adjoint method to inversely design printable microparticles tailored for levitated optomechanics. Our method allows identifying optimized geometries, characterized by enhanced optical trapping and detection efficiencies compared to conventional microspheres. This improves the feasibility of quantum-limited motional control of all translational and rotational degrees of freedom in a standard standing-wave optical trap. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 16:27:00 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 08:40:26 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Lee",
"Moosung",
""
],
[
"Stickler",
"Benjamin A.",
""
],
[
"Pertsch",
"Thomas",
""
],
[
"Hong",
"Sungkun",
""
]
] |
2506.01857 | Boryeu Mao | Boryeu Mao | Protein folding classes -- High-dimensional geometry of amino acid composition space revisited | 45 pages, 5 figures, 4 tables | null | null | null | q-bio.BM q-bio.QM | http://creativecommons.org/licenses/by-nc-sa/4.0/ | In this study, the distributions of protein folding classes of experimentally determined structures from a legacy dataset and a comprehensive database SCOPe are modeled with precise geometric objects as convex polytopes in the high-dimensional amino acid composition space. This is a follow-up of a previous non-statistical, geometry-motivated modeling of protein classes with ellipsoidal models, which are superseded presently in three important respects: (1) as a paradigm shift descriptive 'distribution models' of experimental data are de-coupled from, and serves as the basis for, future potential predictive 'domain models' generalizable to proteins in the same structure class for which the 3-dimensional structures have yet to be determined experimentally, (2) the geometric and analytic characteristics of class distributions are obtained via exact computational geometry calculations, and (3) the full data from a comprehensive database are included in such calculations, eschewing training set selection and biases. In contrast to statistical inference and machine-learning approaches, the analytical, non-statistical geometry models of protein class distributions demonstrated in this study, complete with precise information on the size of class distributions and their relative disposition in the high-dimensional space, and intended not principally for prediction but as accurate description of the complex relationships between amino acid composition and protein classes, suggest that they may ultimately be useful adjuncts for validating sequence-based methods for predicting protein structures and contribute to the mechanistic understanding of secondary structure formation and the folding of polypeptide chains into 3-dimensional conformation of functional protein molecules. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 16:44:02 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Mao",
"Boryeu",
""
]
] |
2506.01870 | Zhi-Wei Sun | Zhi-Wei Sun | Various irrational series involving binomial coefficients | 9 pages. Add more conjectures | null | null | null | math.NT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Motivated by Galois theory, we propose some new irrational series of Ramanujan's type or Zeilberger's type. For example, we conjecture that \begin{align*}&\sum_{k=1}^\infty\frac{(-4)^k(84(1121-338\sqrt{11}))^{4k} (5980ak-b)}{k^3\binom{2k}k^2\binom{4k}{2k}} \\&\qquad\ =36120^2\left(36G-11\sqrt{11}L_{-11}(2)\right), \end{align*} where $$a=11092\sqrt{11}+19437\ \ \text{and}\ \ b=11937508\sqrt{11}+37515813,$$ and $$G=\sum_{k=0}^\infty\frac{(-1)^k}{(2k+1)^2}\ \ \text{and}\ \ L_{-11}(2)=\sum_{k=1}^\infty\frac{(\frac{-11}k)}{k^2}$$ with $(\frac{-11}k)$ the Kronecker symbol. This provides a quite efficient way to compute the constant $L_{-11}(2)$. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 16:56:21 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 16:59:47 GMT"
}
] | 2025-06-05T00:00:00 | [
[
"Sun",
"Zhi-Wei",
""
]
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.