# Research Papers — 2025-10 · bundle page 173 of 217

> Full-text markdown of papers #8601–8650 of 10843 published in 2025-10, newest first. AI-readable mirrors of the full 10843-paper 2025-10 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2025-10/page/173
[← Previous bundle](https://aipod.dev/papers/archive/2025-10/page/172.md)
[Next bundle →](https://aipod.dev/papers/archive/2025-10/page/174.md)

### MASA: Rethinking the Representational Bottleneck in LoRA with Multi-A Shared Adaptation

Authors: Qin Dong, Yuntian Tang, Heming Jia, Yunhang Shen, Bohan Jia, Wenxuan Huang, Lianyue Zhang, Jiao Xie, Shaohui Lin, Rongrong Ji
Published: 2025-10-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.06005v3

Low-Rank Adaptation (LoRA) has emerged as a dominant method in Parameter-Efficient Fine-Tuning (PEFT) for large language models, which augments the transformer layer with one down-projection $A$ and one up-projection $B$. However, LoRA's reliance on a single down-projection matrix ($A$) creates a representational bottleneck, as this solitary feature extractor is inherently insufficient for capturing the diverse signals required by complex tasks. This motivates our architectural shift to focus on enriching the feature adaptation to improve the downstream task adaptation ability. We propose MASA (Multi-$A$ Shared Adaptation), an architecture that implements a multi-$A$, single-$B$ structure where the multi-$A$ expert ensemble is asymmetrically shared across layers to ensure parameter efficiency. In MASA, these specialized experts capture diverse features, which are then integrated by a single, layer-specific $B$-matrix. The effectiveness and versatility of our method are validated through a comprehensive suite of experiments spanning multi-domain generalization, single-domain specialization, and multi-task reasoning. For example, on the MMLU benchmark, MASA achieves an average accuracy of 59.62%, outperforming the standard LoRA by 1.08 points (a relative improvement of 1.84%) with comparable learnable parameters of 0.52%.

---

### Deterministic Legal Agents: A Canonical Primitive API for Auditable Reasoning over Temporal Knowledge Graphs

Authors: Hudson de Martim
Published: 2025-10-07
Categories: cs.AI, cs.CL, cs.IR
arXiv: https://arxiv.org/abs/2510.06002v3

In high-stakes legal domains, retrieval must preserve not only semantic relevance, but also the hierarchy, temporality, and causal provenance of legal norms. Standard Retrieval-Augmented Generation (RAG), based mainly on semantic similarity over text fragments, cannot reliably provide this level of control. Prior work on SAT-Graph RAG addressed the representation problem by modeling legal materials as structure-aware temporal knowledge graphs. This paper addresses the next problem: how an LLM-based reasoning agent can interact with such a graph without reintroducing unreliable retrieval behavior. We specify the SAT-Graph API, a canonical primitive interface for auditable reasoning over temporal knowledge graphs, developed and illustrated in the legal domain. The API exposes typed, atomic, and composable primitives that mediate between a probabilistic language model and a deterministic symbolic substrate. Its design follows Probability Isolation: uncertainty is confined to intent translation, semantic anchoring, and final narrative synthesis, while structural, temporal, and causal graph traversals are executed through deterministic operations over canonical anchors. The interface shifts legal RAG from single-shot Retrieve-then-Generate to active Reason-Act-Observe. An agent decomposes a legal question into an explicit execution plan, invokes primitives for point-in-time retrieval, context reconstruction, provenance tracing, and impact analysis, and produces an answer grounded in an auditable log of graph operations. The result is a formal architectural specification, not an empirical benchmark: a secure interaction protocol that decouples legal knowledge representation from agentic reasoning. Although illustrated in law, the primitive model is domain-portable to other temporally versioned, provenance-sensitive, and authority-governed knowledge bases.

---

### Exploring Gaps in the APS: Direct Minimal Pair Analysis in LLM Syntactic Assessments

Authors: Timothy Pistotti, Jason Brown, Michael Witbrock
Published: 2025-10-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.06001v1

Recent studies probing the Argument from the Poverty of the Stimulus (APS) have applied Large Language Models (LLMs) to test the learnability of complex syntax through surprisal-based metrics. However, divergent conclusions raise questions concerning the insights these metrics offer. While Wilcox et al. (2024) used direct minimal pair comparisons (the "wh-effect") to demonstrate that models successfully generalise knowledge of filler-gap dependencies, Lan et al. (2024) used a Difference-in-Differences (DiD) metric and found that models largely fail on parasitic gaps (PGs). This paper argues that the direct minimal pair approach offers greater diagnostic transparency. We demonstrate this by generating a full 8-permutation paradigm of refined PG stimuli and evaluating the GPT-2 model used in previous studies with a systematic Wilcox-style wh-effect analysis. Our results show that GPT-2 succeeds across all four tested conditions, indicating robust knowledge of filler-gap licensing principles even in complex PG environments. This finding, which contrasts with the more ambiguous results from DiD-style metrics, suggests that the choice of evaluation metric is critical for assessing an LLM's syntactic competence.

---

### Information-Theoretic Policy Pre-Training with Empowerment

Authors: Moritz Schneider, Robert Krug, Narunas Vaskevicius, Luigi Palmieri, Michael Volpp, Joschka Boedecker
Published: 2025-10-07
Categories: cs.AI, cs.IT, cs.LG, cs.RO
arXiv: https://arxiv.org/abs/2510.05996v1

Empowerment, an information-theoretic measure of an agent's potential influence on its environment, has emerged as a powerful intrinsic motivation and exploration framework for reinforcement learning (RL). Besides for unsupervised RL and skill learning algorithms, the specific use of empowerment as a pre-training signal has received limited attention in the literature. We show that empowerment can be used as a pre-training signal for data-efficient downstream task adaptation. For this we extend the traditional notion of empowerment by introducing discounted empowerment, which balances the agent's control over the environment across short- and long-term horizons. Leveraging this formulation, we propose a novel pre-training paradigm that initializes policies to maximize discounted empowerment, enabling agents to acquire a robust understanding of environmental dynamics. We analyze empowerment-based pre-training for various existing RL algorithms and empirically demonstrate its potential as a general-purpose initialization strategy: empowerment-maximizing policies with long horizons are data-efficient and effective, leading to improved adaptability in downstream tasks. Our findings pave the way for future research to scale this framework to high-dimensional and complex tasks, further advancing the field of RL.

---

### Sample Smart, Not Hard: Correctness-First Decoding for Better Reasoning in LLMs

Authors: Xueyan Li, Guinan Su, Mrinmaya Sachan, Jonas Geiping
Published: 2025-10-07
Categories: cs.LG, cs.CL
arXiv: https://arxiv.org/abs/2510.05987v1

Large Language Models (LLMs) are increasingly applied to complex tasks that require extended reasoning. In such settings, models often benefit from diverse chains-of-thought to arrive at multiple candidate solutions. This requires two competing objectives: to inject enough stochasticity to explore multiple reasoning chains, and to ensure sufficient accuracy and quality in each path. Existing works pursue the first objective by increasing exploration at highly uncertain steps with higher temperature or larger candidate token sets, while others improve reliability by rejecting samples with low confidence post-generation, implying that low confidence correlates with low answer quality. These two lines of thought are in conflict, as they conflate different sources of uncertainty. To resolve this, we argue that the decoding rule should be calibrated by correctness, not confidence alone. We should sample from tokens with higher estimated correctness, and reduce sampling where expected correctness is low. We propose simple strategies that achieve this goal: Greedy-Threshold makes sampling greedy at very low confidence steps. Calibrated-TopK and Calibrated-epsilon set truncation threshold based on estimated rank-wise correctness. Together, our findings challenge prevailing heuristics about decoding under uncertainty and show gains across math and general reasoning benchmarks.

---

### ECTSpeech: Enhancing Efficient Speech Synthesis via Easy Consistency Tuning

Authors: Tao Zhu, Yinfeng Yu, Liejun Wang, Fuchun Sun, Wendong Zheng
Published: 2025-10-07
Categories: cs.SD, cs.AI, eess.AS
arXiv: https://arxiv.org/abs/2510.05984v1

Diffusion models have demonstrated remarkable performance in speech synthesis, but typically require multi-step sampling, resulting in low inference efficiency. Recent studies address this issue by distilling diffusion models into consistency models, enabling efficient one-step generation. However, these approaches introduce additional training costs and rely heavily on the performance of pre-trained teacher models. In this paper, we propose ECTSpeech, a simple and effective one-step speech synthesis framework that, for the first time, incorporates the Easy Consistency Tuning (ECT) strategy into speech synthesis. By progressively tightening consistency constraints on a pre-trained diffusion model, ECTSpeech achieves high-quality one-step generation while significantly reducing training complexity. In addition, we design a multi-scale gate module (MSGate) to enhance the denoiser's ability to fuse features at different scales. Experimental results on the LJSpeech dataset demonstrate that ECTSpeech achieves audio quality comparable to state-of-the-art methods under single-step sampling, while substantially reducing the model's training cost and complexity.

---

### Diffusion-Based Image Editing for Breaking Robust Watermarks

Authors: Yunyi Ni, Finn Carter, Ze Niu, Emily Davis, Bo Zhang
Published: 2025-10-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.05978v1

Robust invisible watermarking aims to embed hidden information into images such that the watermark can survive various image manipulations. However, the rise of powerful diffusion-based image generation and editing techniques poses a new threat to these watermarking schemes. In this paper, we present a theoretical study and method demonstrating that diffusion models can effectively break robust image watermarks that were designed to resist conventional perturbations. We show that a diffusion-driven ``image regeneration'' process can erase embedded watermarks while preserving perceptual image content. We further introduce a novel guided diffusion attack that explicitly targets the watermark signal during generation, significantly degrading watermark detectability. Theoretically, we prove that as an image undergoes sufficient diffusion-based transformation, the mutual information between the watermarked image and the embedded watermark payload vanishes, resulting in decoding failure. Experimentally, we evaluate our approach on multiple state-of-the-art watermarking schemes (including the deep learning-based methods StegaStamp, TrustMark, and VINE) and demonstrate near-zero watermark recovery rates after attack, while maintaining high visual fidelity of the regenerated images. Our findings highlight a fundamental vulnerability in current robust watermarking techniques against generative model-based attacks, underscoring the need for new watermarking strategies in the era of generative AI.

---

### A Dynamic Mode Decomposition Approach to Morphological Component Analysis

Authors: Owen T. Huber, Raghu G. Raj, Tianyu Chen, Zacharie I. Idriss
Published: 2025-10-07
Categories: cs.CV, eess.IV
arXiv: https://arxiv.org/abs/2510.05977v1

This paper introduces a novel methodology of adapting the representation of videos based on the dynamics of their scene content variation. In particular, we demonstrate how the clustering of dynamic mode decomposition eigenvalues can be leveraged to learn an adaptive video representation for separating structurally distinct morphologies of a video. We extend the morphological component analysis (MCA) algorithm, which uses multiple predefined incoherent dictionaries and a sparsity prior to separate distinct sources in signals, by introducing our novel eigenspace clustering technique to obtain data-driven MCA dictionaries, which we call dynamic morphological component analysis (DMCA). After deriving our novel algorithm, we offer a motivational example of DMCA applied to a still image, then demonstrate DMCA's effectiveness in denoising applications on videos from the Adobe 240fps dataset. Afterwards, we provide an example of DMCA enhancing the signal-to-noise ratio of a faint target summed with a sea state, and conclude the paper by applying DMCA to separate a bicycle from wind clutter in inverse synthetic aperture radar images.

---

### Diffusion Models for Low-Light Image Enhancement: A Multi-Perspective Taxonomy and Performance Analysis

Authors: Eashan Adhikarla, Yixin Liu, Brian D. Davison
Published: 2025-10-07
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.05976v1

Low-light image enhancement (LLIE) is vital for safety-critical applications such as surveillance, autonomous navigation, and medical imaging, where visibility degradation can impair downstream task performance. Recently, diffusion models have emerged as a promising generative paradigm for LLIE due to their capacity to model complex image distributions via iterative denoising. This survey provides an up-to-date critical analysis of diffusion models for LLIE, distinctively featuring an in-depth comparative performance evaluation against Generative Adversarial Network and Transformer-based state-of-the-art methods, a thorough examination of practical deployment challenges, and a forward-looking perspective on the role of emerging paradigms like foundation models. We propose a multi-perspective taxonomy encompassing six categories: Intrinsic Decomposition, Spectral & Latent, Accelerated, Guided, Multimodal, and Autonomous; that map enhancement methods across physical priors, conditioning schemes, and computational efficiency. Our taxonomy is grounded in a hybrid view of both the model mechanism and the conditioning signals. We evaluate qualitative failure modes, benchmark inconsistencies, and trade-offs between interpretability, generalization, and inference efficiency. We also discuss real-world deployment constraints (e.g., memory, energy use) and ethical considerations. This survey aims to guide the next generation of diffusion-based LLIE research by highlighting trends and surfacing open research questions, including novel conditioning, real-time adaptation, and the potential of foundation models.

---

### LexiCon: a Benchmark for Planning under Temporal Constraints in Natural Language

Authors: Periklis Mantenoglou, Rishi Hazra, Pedro Zuidberg Dos Martires, Luc De Raedt
Published: 2025-10-07
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.05972v1

Owing to their reasoning capabilities, large language models (LLMs) have been evaluated on planning tasks described in natural language. However, LLMs have largely been tested on planning domains without constraints. In order to deploy them in real-world settings where adherence to constraints, in particular safety constraints, is critical, we need to evaluate their performance on constrained planning tasks. We introduce LexiCon -- a natural language-based (Lexi) constrained (Con) planning benchmark, consisting of a suite of environments, that can be used to evaluate the planning capabilities of LLMs in a principled fashion. The core idea behind LexiCon is to take existing planning environments and impose temporal constraints on the states. These constrained problems are then translated into natural language and given to an LLM to solve. A key feature of LexiCon is its extensibility. That is, the set of supported environments can be extended with new (unconstrained) environment generators, for which temporal constraints are constructed automatically. This renders LexiCon future-proof: the hardness of the generated planning problems can be increased as the planning capabilities of LLMs improve. Our experiments reveal that the performance of state-of-the-art LLMs, including reasoning models like GPT-5, o3, and R1, deteriorates as the degree of constrainedness of the planning tasks increases.

---

### Shaken or Stirred? An Analysis of MetaFormer's Token Mixing for Medical Imaging

Authors: Ron Keuth, Paul Kaftan, Mattias P. Heinrich
Published: 2025-10-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.05971v3

The generalization of the Transformer architecture via MetaFormer has reshaped our understanding of its success in computer vision. By replacing self-attention with simpler token mixers, MetaFormer provides strong baselines for vision tasks. However, while extensively studied on natural image datasets, its use in medical imaging remains scarce, and existing works rarely compare different token mixers, potentially overlooking more suitable designs choices. In this work, we present the first comprehensive study of token mixers for medical imaging. We systematically analyze pooling-, convolution-, and attention-based token mixers within the MetaFormer architecture on image classification (global prediction task) and semantic segmentation (dense prediction task). Our evaluation spans nine datasets (seven 2D and two 3D) covering diverse modalities and common challenges in the medical domain. Given the prevalence of pretraining from natural images to mitigate medical data scarcity, we also examine transferring pretrained weights to new token mixers. Our results show that, for classification, low-complexity token mixers (e.g. grouped convolution or pooling) are sufficient, aligning with findings on natural images. Pretrained weights remain useful in some settings despite the domain gap introduced by the new token mixer. For segmentation, we find that the local inductive bias of convolutional token mixers is essential. Grouped convolutions emerge as the preferred choice, as they reduce runtime and parameter count compared to standard convolutions, while the MetaFormer's channel-MLPs already provide the necessary cross-channel interactions.

---

### Probing the Difficulty Perception Mechanism of Large Language Models

Authors: Sunbowen Lee, Qingyu Yin, Chak Tou Leong, Jialiang Zhang, Yicheng Gong, Shiwen Ni, Min Yang, Xiaoyu Shen
Published: 2025-10-07
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.05969v3

Large language models (LLMs) are increasingly deployed on complex reasoning tasks, yet little is known about their ability to internally evaluate problem difficulty, which is an essential capability for adaptive reasoning and efficient resource allocation. In this work, we investigate whether LLMs implicitly encode problem difficulty in their internal representations. Using a linear probe on the final-token representations of LLMs, we demonstrate that the difficulty level of math problems can be linearly modeled. We further locate the specific attention heads of the final Transformer layer: these attention heads have opposite activation patterns for simple and difficult problems, thus achieving perception of difficulty. Our ablation experiments prove the accuracy of the location. Crucially, our experiments provide practical support for using LLMs as automatic difficulty annotators, potentially substantially reducing reliance on costly human labeling in benchmark construction and curriculum learning. We also uncover that there is a significant difference in entropy and difficulty perception at the token level. Our study reveals that difficulty perception in LLMs is not only present but also structurally organized, offering new theoretical insights and practical directions for future research. Our code is available at https://github.com/Aegis1863/Difficulty-Perception-of-LLMs.

---

### MatheMagic: Generating Dynamic Mathematics Benchmarks Robust to Memorization

Authors: Dayyán O'Brien, Barry Haddow, Emily Allaway, Pinzhen Chen
Published: 2025-10-07
Categories: cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2510.05962v1

Conducting contamination-free evaluation of mathematical capabilities can be difficult for two reasons: models may memorize a test set once it is made public, and current mathematical benchmarks are prone to overfitting due to having limited diversity of symbols and rules, coupled with closed-ended answers. This paper proposes a method to leverage these shortcomings as useful features to a construct dynamic, counterfactual benchmark, which can be used to both reveal overfitting and measure true reasoning. We demonstrate this via MatheMagic, which generates math test instances with the interpretations of numbers and operators altered, yet has automatically verifiable answers. Test instances are randomly seeded and constructed at test time to evaluate a model's induction or deduction capability, offering stability, extensibility, comparability, and robustness to overfitting. Our experiments find that models solve deduction more easily than induction, but they revert to standard math. Further analysis reveals that math-adapted models fail to exhibit a general "skill" of reasoning, and fine-tuning on induction tasks generalizes poorly.

---

### Training-Free Time Series Classification via In-Context Reasoning with LLM Agents

Authors: Songyuan Sui, Zihang Xu, Xia Hu
Published: 2025-10-07
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.05950v2

Time series classification (TSC) spans diverse application scenarios, yet labeled data are often scarce, making task-specific training costly and inflexible. Recent reasoning-oriented large language models (LLMs) show promise in understanding temporal patterns, but purely zero-shot usage remains suboptimal. We propose FETA, a multi-agent framework for training-free TSC via exemplar-based in-context reasoning. FETA decomposes a multivariate series into channel-wise subproblems, retrieves a few structurally similar labeled examples for each channel, and leverages a reasoning LLM to compare the query against these exemplars, producing channel-level labels with self-assessed confidences; a confidence-weighted aggregator then fuses all channel decisions. This design eliminates the need for pretraining or fine-tuning, improves efficiency by pruning irrelevant channels and controlling input length, and enhances interpretability through exemplar grounding and confidence estimation. On nine challenging UEA datasets, FETA achieves strong accuracy under a fully training-free setting, surpassing multiple trained baselines. These results demonstrate that a multi-agent in-context reasoning framework can transform LLMs into competitive, plug-and-play TSC solvers without any parameter training. The code is available at https://github.com/SongyuanSui/FETATSC.

---

### Gaussian Embeddings: How JEPAs Secretly Learn Your Data Density

Authors: Randall Balestriero, Nicolas Ballas, Mike Rabbat, Yann LeCun
Published: 2025-10-07
Categories: cs.LG, cs.AI, cs.CV, stat.ML
arXiv: https://arxiv.org/abs/2510.05949v1

Joint Embedding Predictive Architectures (JEPAs) learn representations able to solve numerous downstream tasks out-of-the-box. JEPAs combine two objectives: (i) a latent-space prediction term, i.e., the representation of a slightly perturbed sample must be predictable from the original sample's representation, and (ii) an anti-collapse term, i.e., not all samples should have the same representation. While (ii) is often considered as an obvious remedy to representation collapse, we uncover that JEPAs' anti-collapse term does much more--it provably estimates the data density. In short, any successfully trained JEPA can be used to get sample probabilities, e.g., for data curation, outlier detection, or simply for density estimation. Our theoretical finding is agnostic of the dataset and architecture used--in any case one can compute the learned probabilities of sample $x$ efficiently and in closed-form using the model's Jacobian matrix at $x$. Our findings are empirically validated across datasets (synthetic, controlled, and Imagenet) and across different Self Supervised Learning methods falling under the JEPA family (I-JEPA and DINOv2) and on multimodal models, such as MetaCLIP. We denote the method extracting the JEPA learned density as {\bf JEPA-SCORE}.

---

### N-Parties Private Structure and Parameter Learning for Sum-Product Networks

Authors: Xenia Heilmann, Ernst Althaus, Mattia Cerrato, Nick Johannes Peter Rassau, Mohammad Sadeq Dousti, Stefan Kramer
Published: 2025-10-07
Categories: cs.CR, cs.LG
arXiv: https://arxiv.org/abs/2510.05946v1

A sum-product network (SPN) is a graphical model that allows several types of probabilistic inference to be performed efficiently. In this paper, we propose a privacy-preserving protocol which tackles structure generation and parameter learning of SPNs. Additionally, we provide a protocol for private inference on SPNs, subsequent to training. To preserve the privacy of the participants, we derive our protocol based on secret sharing, which guarantees privacy in the honest-but-curious setting even when at most half of the parties cooperate to disclose the data. The protocol makes use of a forest of randomly generated SPNs, which is trained and weighted privately and can then be used for private inference on data points. Our experiments indicate that preserving the privacy of all participants does not decrease log-likelihood performance on both homogeneously and heterogeneously partitioned data. We furthermore show that our protocol's performance is comparable to current state-of-the-art SPN learners in homogeneously partitioned data settings. In terms of runtime and memory usage, we demonstrate that our implementation scales well when increasing the number of parties, comparing favorably to protocols for neural networks, when they are trained to reproduce the input-output behavior of SPNs.

---

### EARL: Efficient Agentic Reinforcement Learning Systems for Large Language Models

Authors: Zheyue Tan, Mustapha Abdullahi, Tuo Shi, Huining Yuan, Zelai Xu, Chao Yu, Boxun Li, Bo Zhao
Published: 2025-10-07
Categories: cs.DC, cs.LG
arXiv: https://arxiv.org/abs/2510.05943v1

Reinforcement learning (RL) has become a pivotal component of large language model (LLM) post-training, and agentic RL extends this paradigm to operate as agents through multi-turn interaction and tool use. Scaling such systems exposes two practical bottlenecks: (1) context length grows rapidly during training, inflating memory usage and latency, and triggering out-of-memory (OOM) failures; and (2) intermediate tensors accumulate with context length, making cross-device data movement a major system bottleneck. We present EARL, a scalable system for efficient agentic RL. EARL designs a parallelism selector that dynamically adapts model and training parallelism across RL stages based on sequence length and system load, and a data dispatcher that performs layout-aware, decentralized exchange of intermediate data batches. Together, these components increase throughput, reduce long-context failures, and enable stable large-scale training of agentic LLMs without relying on hard limits or penalties of context length.

---

### EvalMORAAL: Interpretable Chain-of-Thought and LLM-as-Judge Evaluation for Moral Alignment in Large Language Models

Authors: Hadi Mohammadi, Anastasia Giachanou, Robert A. Bagheri
Published: 2025-10-07
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.05942v3

We present EvalMORAAL, a transparent chain-of-thought (CoT) framework that uses two scoring methods (log-probabilities and direct ratings) plus a model-as-judge peer review to evaluate moral alignment in 20 large language models. We assess models on the World Values Survey (55 countries, 19 topics) and the PEW Global Attitudes Survey (39 countries, 8 topics). With EvalMORAAL, top models align closely with survey responses (Pearson's $r \approx 0.90$ on WVS). Yet we find a clear regional difference: Western regions average $r=0.82$ while non-Western regions average $r=0.61$ (a 0.21 absolute gap), indicating a persistent regional alignment gap. Our framework adds three parts: (1) two scoring methods for all models to enable fair comparison, (2) a structured CoT protocol with self-consistency checks, and (3) a model-as-judge peer review that flags 348 conflicts using a data-driven threshold. Peer agreement relates to WVS survey alignment ($r=0.74$, $p<.001$; PEW $r=0.39$, n.s.), supporting automated quality checks. These results show real progress toward culture-aware AI while highlighting open challenges for use across regions.

---

### LLM-FS-Agent: A Deliberative Role-based Large Language Model Architecture for Transparent Feature Selection

Authors: Mohamed Bal-Ghaoui, Fayssal Sabri
Published: 2025-10-07
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.05935v1

High-dimensional data remains a pervasive challenge in machine learning, often undermining model interpretability and computational efficiency. While Large Language Models (LLMs) have shown promise for dimensionality reduction through feature selection, existing LLM-based approaches frequently lack structured reasoning and transparent justification for their decisions. This paper introduces LLM-FS-Agent, a novel multi-agent architecture designed for interpretable and robust feature selection. The system orchestrates a deliberative "debate" among multiple LLM agents, each assigned a specific role, enabling collective evaluation of feature relevance and generation of detailed justifications. We evaluate LLM-FS-Agent in the cybersecurity domain using the CIC-DIAD 2024 IoT intrusion detection dataset and compare its performance against strong baselines, including LLM-Select and traditional methods such as PCA. Experimental results demonstrate that LLM-FS-Agent consistently achieves superior or comparable classification performance while reducing downstream training time by an average of 46% (statistically significant improvement, p = 0.028 for XGBoost). These findings highlight that the proposed deliberative architecture enhances both decision transparency and computational efficiency, establishing LLM-FS-Agent as a practical and reliable solution for real-world applications.

---

### Hire Your Anthropologist! Rethinking Culture Benchmarks Through an Anthropological Lens

Authors: Mai AlKhamissi, Yunze Xiao, Badr AlKhamissi, Mona Diab
Published: 2025-10-07
Categories: cs.CL, cs.CY
arXiv: https://arxiv.org/abs/2510.05931v2

Cultural evaluation of large language models has become increasingly important, yet current benchmarks often reduce culture to static facts or homogeneous values. This view conflicts with anthropological accounts that emphasize culture as dynamic, historically situated, and enacted in practice. To analyze this gap, we introduce a four-part framework that categorizes how benchmarks frame culture, such as knowledge, preference, performance, or bias. Using this lens, we qualitatively examine 20 cultural benchmarks and identify six recurring methodological issues, including treating countries as cultures, overlooking within-culture diversity, and relying on oversimplified survey formats. Drawing on established anthropological methods, we propose concrete improvements: incorporating real-world narratives and scenarios, involving cultural communities in design and validation, and evaluating models in context rather than isolation. Our aim is to guide the development of cultural benchmarks that go beyond static recall tasks and more accurately capture the responses of the models to complex cultural situations.

---

### Carré du champ flow matching: better quality-generalisation tradeoff in generative models

Authors: Jacob Bamberger, Iolo Jones, Dennis Duncan, Michael M. Bronstein, Pierre Vandergheynst, Adam Gosztolai
Published: 2025-10-07
Categories: cs.LG, cs.AI, math.DG
arXiv: https://arxiv.org/abs/2510.05930v2

Deep generative models often face a fundamental tradeoff: high sample quality can come at the cost of memorisation, where the model reproduces training data rather than generalising across the underlying data geometry. We introduce Carré du champ flow matching (CDC-FM), a generalisation of flow matching (FM), that improves the quality-generalisation tradeoff by regularising the probability path with a geometry-aware noise. Our method replaces the homogeneous, isotropic noise in FM with a spatially varying, anisotropic Gaussian noise whose covariance captures the local geometry of the latent data manifold. We prove that this geometric noise can be optimally estimated from the data and is scalable to large data. Further, we provide an extensive experimental evaluation on diverse datasets (synthetic manifolds, point clouds, single-cell genomics, animal motion capture, and images) as well as various neural network architectures (MLPs, CNNs, and transformers). We demonstrate that CDC-FM consistently offers a better quality-generalisation tradeoff. We observe significant improvements over standard FM in data-scarce regimes and in highly non-uniformly sampled datasets, which are often encountered in AI for science applications. Our work provides a mathematical framework for studying the interplay between data geometry, generalisation and memorisation in generative models, as well as a robust and scalable algorithm that can be readily integrated into existing flow matching pipelines.

---

### A Warm-basis Method for Bridging Learning and Iteration: a Case Study in Fluorescence Molecular Tomography

Authors: Ruchi Guo, Jiahua Jiang, Bangti Jin, Wuwei Ren, Jianru Zhang
Published: 2025-10-07
Categories: math.NA, cs.CV
arXiv: https://arxiv.org/abs/2510.05926v1

Fluorescence Molecular Tomography (FMT) is a widely used non-invasive optical imaging technology in biomedical research. It usually faces significant accuracy challenges in depth reconstruction, and conventional iterative methods struggle with poor $z$-resolution even with advanced regularization. Supervised learning approaches can improve recovery accuracy but rely on large, high-quality paired training dataset that is often impractical to acquire in practice. This naturally raises the question of how learning-based approaches can be effectively combined with iterative schemes to yield more accurate and stable algorithms. In this work, we present a novel warm-basis iterative projection method (WB-IPM) and establish its theoretical underpinnings. The method is able to achieve significantly more accurate reconstructions than the learning-based and iterative-based methods. In addition, it allows a weaker loss function depending solely on the directional component of the difference between ground truth and neural network output, thereby substantially reducing the training effort. These features are justified by our error analysis as well as simulated and real-data experiments.

---

### Prompt reinforcing for long-term planning of large language models

Authors: Hsien-Chin Lin, Benjamin Matthias Ruppik, Carel van Niekerk, Chia-Hao Shen, Michael Heck, Nurul Lubis, Renato Vukovic, Shutong Feng, Milica Gašić
Published: 2025-10-07
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2510.05921v3

Large language models (LLMs) have achieved remarkable success in a wide range of natural language processing tasks and can be adapted through prompting. However, they remain suboptimal in multi-turn interactions, often relying on incorrect early assumptions and failing to track user goals over time, which makes such tasks particularly challenging. Prior works in dialogue systems have shown that long-term planning is essential for handling interactive tasks. In this work, we propose a prompt optimisation framework inspired by reinforcement learning, which enables such planning to take place by only modifying the task instruction prompt of the LLM-based agent. By generating turn-by-turn feedback and leveraging experience replay for prompt rewriting, our proposed method shows significant improvement in multi-turn tasks such as text-to-SQL and task-oriented dialogue. Moreover, it generalises across different LLM-based agents and can leverage diverse LLMs as meta-prompting agents. This warrants future research in reinforcement learning-inspired parameter-free optimisation methods.

---

### An Attention-Augmented VAE-BiLSTM Framework for Anomaly Detection in 12-Lead ECG Signals

Authors: Marc Garreta Basora, Mehmet Oguz Mulayim
Published: 2025-10-07
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.05919v1

Anomaly detection in 12-lead electrocardiograms (ECGs) is critical for identifying deviations associated with cardiovascular disease. This work presents a comparative analysis of three autoencoder-based architectures: convolutional autoencoder (CAE), variational autoencoder with bidirectional long short-term memory (VAE-BiLSTM), and VAE-BiLSTM with multi-head attention (VAE-BiLSTM-MHA), for unsupervised anomaly detection in ECGs. To the best of our knowledge, this study reports the first application of a VAE-BiLSTM-MHA architecture to ECG anomaly detection. All models are trained on normal ECG samples to reconstruct non-anomalous cardiac morphology and detect deviations indicative of disease. Using a unified preprocessing and evaluation pipeline on the public China Physiological Signal Challenge (CPSC) dataset, the attention-augmented VAE achieves the best performance, with an AUPRC of 0.81 and a recall of 0.85 on the held-out test set, outperforming the other architectures. To support clinical triage, this model is further integrated into an interactive dashboard that visualizes anomaly localization. In addition, a performance comparison with baseline models from the literature is provided.

---

### Optimizing for Persuasion Improves LLM Generalization: Evidence from Quality-Diversity Evolution of Debate Strategies

Authors: Aksel Joonas Reedi, Corentin Léger, Julien Pourcel, Loris Gaven, Perrine Charriau, Guillaume Pourcel
Published: 2025-10-07
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.05909v2

Large Language Models (LLMs) optimized to output truthful answers often overfit, producing brittle reasoning that fails to generalize. While persuasion-based optimization has shown promise in debate settings, it has not been systematically compared against mainstream truth-based approaches. We introduce DebateQD, a minimal Quality-Diversity (QD) evolutionary algorithm that evolves diverse debate strategies across different categories (rationality, authority, emotional appeal, etc.) through tournament-style competitions where two LLMs debate while a third judges. Unlike previously proposed methods that require a population of LLMs, our approach maintains diversity of opponents through prompt-based strategies within a single LLM architecture, making it more accessible for experiments while preserving the key benefits of population-based optimization. In contrast to prior work, we explicitly isolate the role of the optimization objective by fixing the debate protocol and swapping only the fitness function: persuasion rewards strategies that convince the judge irrespective of truth, whereas truth rewards collaborative correctness. Across three model scales (7B, 32B, 72B parameters) and multiple dataset sizes from the QuALITY benchmark, persuasion-optimized strategies achieve up to 13.94% smaller train-test generalization gaps, while matching or exceeding truth optimization's test performance. These results provide the first controlled evidence that competitive pressure to persuade, rather than seek the truth collaboratively, fosters more transferable reasoning skills, offering a promising path for improving LLM generalization.

---

### Kaputt: A Large-Scale Dataset for Visual Defect Detection

Authors: Sebastian Höfer, Dorian Henning, Artemij Amiranashvili, Douglas Morrison, Mariliza Tzes, Ingmar Posner, Marc Matvienko, Alessandro Rennola, Anton Milan
Published: 2025-10-07
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.05903v1

We present a novel large-scale dataset for defect detection in a logistics setting. Recent work on industrial anomaly detection has primarily focused on manufacturing scenarios with highly controlled poses and a limited number of object categories. Existing benchmarks like MVTec-AD (6) and VisA (33) have reached saturation, with state-of-the-art methods achieving up to 99.9% AUROC scores. In contrast to manufacturing, anomaly detection in retail logistics faces new challenges, particularly in the diversity and variability of object pose and appearance. Leading anomaly detection methods fall short when applied to this new setting. To bridge this gap, we introduce a new benchmark that overcomes the current limitations of existing datasets. With over 230,000 images (and more than 29,000 defective instances), it is 40 times larger than MVTec-AD and contains more than 48,000 distinct objects. To validate the difficulty of the problem, we conduct an extensive evaluation of multiple state-of-the-art anomaly detection methods, demonstrating that they do not surpass 56.96% AUROC on our dataset. Further qualitative analysis confirms that existing methods struggle to leverage normal samples under heavy pose and appearance variation. With our large-scale dataset, we set a new benchmark and encourage future research towards solving this challenging problem in retail logistics anomaly detection. The dataset is available for download under https://www.kaputt-dataset.com.

---

### Untangling Component Imbalance in Hybrid Linear Attention Conversion Methods

Authors: Martin Benfeghoul, Teresa Delgado, Adnan Oomerjee, Haitham Bou Ammar, Jun Wang, Zafeirios Fountas
Published: 2025-10-07
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2510.05901v2

Transformers' quadratic computational complexity limits their scalability despite remarkable performance. While linear attention reduces this to linear complexity, pre-training such models from scratch remains, in most cases, prohibitively expensive. Recent post-training linearisation methods convert pre-trained Transformers to linear models efficiently, often using hybrid approaches that combine linear attention with sliding-window softmax. We identify a critical flaw: existing hybrid methods inadvertently bypass the linear component, relying almost entirely on SWA. Component-level diagnostics reveal this previously undetected behaviour stems from overlooked evaluation practices on common-sense benchmarks. We propose three solutions to ensure balanced component usage: (i) inference-time hybridisation of linear-only conversions with sliding-window softmax; (ii) HedgeCATs, combining attention-weight transfer with targeted LoRA fine-tuning; and (iii) Scheduled Sliding-window Dropout (SSD), which stochastically suppresses the softmax branch during training to prevent component collapse. Our methods maintain computational efficiency while recovering most base model performance and ensuring genuine linear attention adoption, restoring the validity of performance attributions in hybrid conversions.

---

### Efficient Universal Models for Medical Image Segmentation via Weakly Supervised In-Context Learning

Authors: Jiesi Hu, Yanwu Yang, Zhiyu Ye, Jinyan Zhou, Jianfeng Cao, Hanyang Peng, Ting Ma
Published: 2025-10-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.05899v2

Universal models for medical image segmentation, such as interactive and in-context learning (ICL) models, offer strong generalization but require extensive annotations. Interactive models need repeated user prompts for each image, while ICL relies on dense, pixel-level labels. To address this, we propose Weakly Supervised In-Context Learning (WS-ICL), a new ICL paradigm that leverages weak prompts (e.g., bounding boxes or points) instead of dense labels for context. This approach significantly reduces annotation effort by eliminating the need for fine-grained masks and repeated user prompting for all images. We evaluated the proposed WS-ICL model on three held-out benchmarks. Experimental results demonstrate that WS-ICL achieves performance comparable to regular ICL models at a significantly lower annotation cost. In addition, WS-ICL is highly competitive even under the interactive paradigm. These findings establish WS-ICL as a promising step toward more efficient and unified universal models for medical image segmentation. Our code and model are publicly available at https://github.com/jiesihu/Weak-ICL.

---

### $\bf{D^3}$QE: Learning Discrete Distribution Discrepancy-aware Quantization Error for Autoregressive-Generated Image Detection

Authors: Yanran Zhang, Bingyao Yu, Yu Zheng, Wenzhao Zheng, Yueqi Duan, Lei Chen, Jie Zhou, Jiwen Lu
Published: 2025-10-07
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2510.05891v1

The emergence of visual autoregressive (AR) models has revolutionized image generation while presenting new challenges for synthetic image detection. Unlike previous GAN or diffusion-based methods, AR models generate images through discrete token prediction, exhibiting both marked improvements in image synthesis quality and unique characteristics in their vector-quantized representations. In this paper, we propose to leverage Discrete Distribution Discrepancy-aware Quantization Error (D$^3$QE) for autoregressive-generated image detection that exploits the distinctive patterns and the frequency distribution bias of the codebook existing in real and fake images. We introduce a discrete distribution discrepancy-aware transformer that integrates dynamic codebook frequency statistics into its attention mechanism, fusing semantic features and quantization error latent. To evaluate our method, we construct a comprehensive dataset termed ARForensics covering 7 mainstream visual AR models. Experiments demonstrate superior detection accuracy and strong generalization of D$^3$QE across different AR models, with robustness to real-world perturbations. Code is available at \href{https://github.com/Zhangyr2022/D3QE}{https://github.com/Zhangyr2022/D3QE}.

---

### BioAutoML-NAS: An End-to-End AutoML Framework for Multimodal Insect Classification via Neural Architecture Search on Large-Scale Biodiversity Data

Authors: Arefin Ittesafun Abian, Debopom Sutradhar, Md Rafi Ur Rashid, Reem E. Mohamed, Md Rafiqul Islam, Asif Karim, Kheng Cher Yeo, Sami Azam
Published: 2025-10-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.05888v2

Insect classification is important for agricultural management and ecological research, as it directly affects crop health and production. However, this task remains challenging due to the complex characteristics of insects, class imbalance, and large-scale datasets. To address these issues, we propose BioAutoML-NAS, the first BioAutoML model using multimodal data, including images, and metadata, which applies neural architecture search (NAS) for images to automatically learn the best operations for each connection within each cell. Multiple cells are stacked to form the full network, each extracting detailed image feature representations. A multimodal fusion module combines image embeddings with metadata, allowing the model to use both visual and categorical biological information to classify insects. An alternating bi-level optimization training strategy jointly updates network weights and architecture parameters, while zero operations remove less important connections, producing sparse, efficient, and high-performing architectures. Extensive evaluation on the BIOSCAN-5M dataset demonstrates that BioAutoML-NAS achieves 96.81% accuracy, 97.46% precision, 96.81% recall, and a 97.05% F1 score, outperforming state-of-the-art transfer learning, transformer, AutoML, and NAS methods by approximately 16%, 10%, and 8% respectively. Further validation on the Insects-1M dataset obtains 93.25% accuracy, 93.71% precision, 92.74% recall, and a 93.22% F1 score. These results demonstrate that BioAutoML-NAS provides accurate, confident insect classification that supports modern sustainable farming.

---

### acia-workflows: Automated Single-cell Imaging Analysis for Scalable and Deep Learning-based Live-cell Imaging Analysis Workflows

Authors: Johannes Seiffarth, Keitaro Kasahara, Michelle Bund, Benita Lückel, Richard D. Paul, Matthias Pesch, Lennart Witting, Michael Bott, Dietrich Kohlheyer, Katharina Nöh
Published: 2025-10-07
Categories: cs.CV, q-bio.QM
arXiv: https://arxiv.org/abs/2510.05886v2

Live-cell imaging (LCI) technology enables the detailed spatio-temporal characterization of living cells at the single-cell level, which is critical for advancing research in the life sciences, from biomedical applications to bioprocessing. High-throughput setups with tens to hundreds of parallel cell cultivations offer the potential for robust and reproducible insights. However, these insights are obscured by the large amount of LCI data recorded per experiment. Recent advances in state-of-the-art deep learning methods for cell segmentation and tracking now enable the automated analysis of such large data volumes, offering unprecedented opportunities to systematically study single-cell dynamics. The next key challenge lies in integrating these powerful tools into accessible, flexible, and user-friendly workflows that support routine application in biological research. In this work, we present acia-workflows, a platform that combines three key components: (1) the Automated live-Cell Imaging Analysis (acia) Python library, which supports the modular design of image analysis pipelines offering eight deep learning segmentation and tracking approaches; (2) workflows that assemble the image analysis pipeline, its software dependencies, documentation, and visualizations into a single Jupyter Notebook, leading to accessible, reproducible and scalable analysis workflows; and (3) a collection of application workflows showcasing the analysis and customization capabilities in real-world applications. Specifically, we present three workflows to investigate various types of microfluidic LCI experiments ranging from growth rate comparisons to precise, minute-resolution quantitative analyses of individual dynamic cells responses to changing oxygen conditions. Our collection of more than ten application workflows is open source and publicly available at https://github.com/JuBiotech/acia-workflows.

---

### Segment-Factorized Full-Song Generation on Symbolic Piano Music

Authors: Ping-Yi Chen, Chih-Pin Tan, Yi-Hsuan Yang
Published: 2025-10-07
Categories: cs.SD, cs.AI, cs.LG, cs.MM, eess.AS
arXiv: https://arxiv.org/abs/2510.05881v1

We propose the Segmented Full-Song Model (SFS) for symbolic full-song generation. The model accepts a user-provided song structure and an optional short seed segment that anchors the main idea around which the song is developed. By factorizing a song into segments and generating each one through selective attention to related segments, the model achieves higher quality and efficiency compared to prior work. To demonstrate its suitability for human-AI interaction, we further wrap SFS into a web application that enables users to iteratively co-create music on a piano roll with customizable structures and flexible ordering.

---

### OBSR: Open Benchmark for Spatial Representations

Authors: Julia Moska, Oleksii Furman, Kacper Kozaczko, Szymon Leszkiewicz, Jakub Polczyk, Piotr Gramacki, Piotr Szymański
Published: 2025-10-07
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.05879v2

GeoAI is evolving rapidly, fueled by diverse geospatial datasets like traffic patterns, environmental data, and crowdsourced OpenStreetMap (OSM) information. While sophisticated AI models are being developed, existing benchmarks are often concentrated on single tasks and restricted to a single modality. As such, progress in GeoAI is limited by the lack of a standardized, multi-task, modality-agnostic benchmark for their systematic evaluation. This paper introduces a novel benchmark designed to assess the performance, accuracy, and efficiency of geospatial embedders. Our benchmark is modality-agnostic and comprises 7 distinct datasets from diverse cities across three continents, ensuring generalizability and mitigating demographic biases. It allows for the evaluation of GeoAI embedders on various phenomena that exhibit underlying geographic processes. Furthermore, we establish a simple and intuitive task-oriented model baselines, providing a crucial reference point for comparing more complex solutions.

---

### MaNGO - Adaptable Graph Network Simulators via Meta-Learning

Authors: Philipp Dahlinger, Tai Hoang, Denis Blessing, Niklas Freymuth, Gerhard Neumann
Published: 2025-10-07
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.05874v2

Accurately simulating physics is crucial across scientific domains, with applications spanning from robotics to materials science. While traditional mesh-based simulations are precise, they are often computationally expensive and require knowledge of physical parameters, such as material properties. In contrast, data-driven approaches like Graph Network Simulators (GNSs) offer faster inference but suffer from two key limitations: Firstly, they must be retrained from scratch for even minor variations in physical parameters, and secondly they require labor-intensive data collection for each new parameter setting. This is inefficient, as simulations with varying parameters often share a common underlying latent structure. In this work, we address these challenges by learning this shared structure through meta-learning, enabling fast adaptation to new physical parameters without retraining. To this end, we propose a novel architecture that generates a latent representation by encoding graph trajectories using conditional neural processes (CNPs). To mitigate error accumulation over time, we combine CNPs with a novel neural operator architecture. We validate our approach, Meta Neural Graph Operator (MaNGO), on several dynamics prediction tasks with varying material properties, demonstrating superior performance over existing GNS methods. Notably, MaNGO achieves accuracy on unseen material properties close to that of an oracle model.

---

### Towards Label-Free Biological Reasoning Synthetic Dataset Creation via Uncertainty Filtering

Authors: Josefa Lia Stoisser, Lawrence Phillips, Aditya Misra, Tom A. Lamb, Philip Torr, Marc Boubnovski Martell, Julien Fauqueur, Kaspar Märtens
Published: 2025-10-07
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.05871v1

Synthetic chain-of-thought (CoT) traces are widely used to train large reasoning models (LRMs), improving generalization by providing step-level supervision. Yet most approaches require ground-truth labels to seed or filter these traces - an expensive bottleneck in domains like biology where wet-lab data are scarce. We propose a label-free alternative: uncertainty-based filtering, which uses a model's own confidence - quantified through established uncertainty metrics like self-consistency and predictive perplexity - as a substitute for external labels. We sample multiple reasoning traces and retain only low-uncertainty subsets. Applied to biological perturbation prediction, a domain where wet-lab labels are especially costly, we show that the filtered subset has higher accuracy, and that supervised fine-tuning (SFT) on uncertainty-filtered data outperforms unfiltered synthetic data, narrows the gap to ground-truth training, and surpasses strong LRM baselines. Ablations show that per-class filtering corrects for class-specific uncertainty scales and that hybrid uncertainty metrics yield higher-quality datasets. Our results suggest that model-internal confidence is a powerful signal for efficient reasoning dataset creation, enabling LRMs in domains where supervision is expensive.

---

### The fragility of "cultural tendencies" in LLMs

Authors: Kun Sun, Rong Wang
Published: 2025-10-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.05869v1

In a recent study, Lu, Song, and Zhang (2025) (LSZ) propose that large language models (LLMs), when prompted in different languages, display culturally specific tendencies. They report that the two models (i.e., GPT and ERNIE) respond in more interdependent and holistic ways when prompted in Chinese, and more independent and analytic ways when prompted in English. LSZ attribute these differences to deep-seated cultural patterns in the models, claiming that prompt language alone can induce substantial cultural shifts. While we acknowledge the empirical patterns they observed, we find their experiments, methods, and interpretations problematic. In this paper, we critically re-evaluate the methodology, theoretical framing, and conclusions of LSZ. We argue that the reported "cultural tendencies" are not stable traits but fragile artifacts of specific models and task design. To test this, we conducted targeted replications using a broader set of LLMs and a larger number of test items. Our results show that prompt language has minimal effect on outputs, challenging LSZ's claim that these models encode grounded cultural beliefs.

---

### The Safety Challenge of World Models for Embodied AI Agents: A Review

Authors: Lorenzo Baraldi, Zifan Zeng, Chongzhe Zhang, Aradhana Nayak, Hongbo Zhu, Feng Liu, Qunli Zhang, Peng Wang, Shiming Liu, Zheng Hu, Angelo Cangelosi, Lorenzo Baraldi
Published: 2025-10-07
Categories: cs.AI, cs.CV, cs.RO
arXiv: https://arxiv.org/abs/2510.05865v1

The rapid progress in embodied artificial intelligence has highlighted the necessity for more advanced and integrated models that can perceive, interpret, and predict environmental dynamics. In this context, World Models (WMs) have been introduced to provide embodied agents with the abilities to anticipate future environmental states and fill in knowledge gaps, thereby enhancing agents' ability to plan and execute actions. However, when dealing with embodied agents it is fundamental to ensure that predictions are safe for both the agent and the environment. In this article, we conduct a comprehensive literature review of World Models in the domains of autonomous driving and robotics, with a specific focus on the safety implications of scene and control generation tasks. Our review is complemented by an empirical analysis, wherein we collect and examine predictions from state-of-the-art models, identify and categorize common faults (herein referred to as pathologies), and provide a quantitative evaluation of the results.

---

### On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs

Authors: Faeze Ghorbanpour, Alexander Fraser
Published: 2025-10-07
Categories: cs.CL, cs.CY
arXiv: https://arxiv.org/abs/2510.05864v2

Large language models (LLMs) increasingly operate on long inputs, yet their behavior when harmful sentences are sparsely embedded within such inputs remains poorly understood. We present a sensitivity analysis that probes how LLMs extract harmful sentences embedded in long inputs. We construct long inputs by combining neutral and harmful sentences, and systematically vary four factors: input length (600--30,000 tokens), the proportion of harmful sentences (0.01--0.50), harm realization (explicit vs. implicit), and the position of harmful sentences within the input (beginning, middle, end), enabling a controlled stress-test evaluation. Experiments across toxic, offensive, and hate content, and across LLaMA-3.1, Qwen-2.5, and Mistral, reveal consistent patterns: sensitivity is non-monotonic with respect to harmful prevalence, peaking at moderate levels; sensitivity degrades as input length increases; harmful sentences placed earlier in the input are more strongly prioritized; and explicit harm is more reliably identified than implicit harm. These findings provide a systematic view of how LLMs prioritize harmful sentences in long input under controlled stress conditions, highlighting both emerging strengths and remaining challenges for safety-related use.

---

### Revisiting Long-context Modeling from Context Denoising Perspective

Authors: Zecheng Tang, Baibei Ji, Juntao Li, Lijun Wu, Haijia Gui, Min Zhang
Published: 2025-10-07
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.05862v2

Long-context models (LCMs) have demonstrated great potential in processing long sequences, facilitating many real-world applications. The success of LCMs can be attributed to their ability to locate implicit critical information within the context for further prediction. However, recent research reveals that LCMs are often susceptible to contextual noise, i.e., irrelevant tokens, that can mislead model attention. In this paper, we conduct a fine-grained analysis of the context noise and propose an effective metric, the Integrated Gradient (IG) score, to detect and quantify the noise information within the context. Our findings reveal that even simple mitigation of detected context noise can substantially boost the model's attention on critical tokens and benefit subsequent predictions. Building on this insight, we propose Context Denoising Training (CDT), a straightforward yet effective training strategy that improves attention on critical tokens while reinforcing their influence on model predictions. Extensive experiments across four tasks, under both context window scaling and long-context alignment settings, demonstrate the superiority of CDT. Notably, when trained with CDT, an open-source 8B model can achieve performance (50.92) comparable to GPT-4o (51.00).

---

### Overcoming Language Barriers: Multilingual Analysis of the 2023 Swiss Privacy Law's Impact

Authors: Luka Nenadic, David Rodriguez, Joseph A. Calandrino
Published: 2025-10-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.05860v2

Policymakers enact and revise privacy laws expecting meaningful benefits for their people in practice. While scholarship has measured the real-world impact of some privacy regulations-the EU and California most notably-limited empirical evidence exists for many of the more than 140 countries that have implemented some form of privacy legislation. Switzerland, a multilingual country bordered almost entirely by EU states, is one such example. This paper analyzes the extent to which a 2023 alignment of Swiss privacy law with EU privacy regulation affected website privacy policies in Switzerland. To address Switzerland's unique multilingual culture, we develop an LLM-based pipeline that extracts legally relevant information as document-level labels in a single inference without requiring translation. On a benchmark of 120 expert-annotated privacy policies in German, French, Italian, and English, our pipeline achieves F1 scores above 0.90 for most pairs of languages and legally relevant disclosures. Applying this pipeline to privacy policies we collected from more than 35,000 Swiss- and EU-facing websites before and after the 2023 privacy law revision, we find significant increases in both mandatory and voluntary disclosures of data subject rights among Swiss privacy policies. In exploring the mechanisms driving increased disclosure rates, we discover heavy use of automated privacy policy generators and find that generated policies are associated with up to 15 percentage points higher disclosure rates. These results provide large-scale empirical evidence of how regulatory change and novel drafting technologies impact the content of privacy policies in a unique multilingual environment.

---

### DACP: Domain-Adaptive Continual Pre-Training of Large Language Models for Phone Conversation Summarization

Authors: Xue-Yong Fu, Elena Khasanova, Md Tahmid Rahman Laskar, Harsh Saini, Shashi Bhushan TN
Published: 2025-10-07
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.05858v3

Large language models (LLMs) have achieved impressive performance in text summarization, yet their performance often falls short when applied to specialized domains that differ from their original pre-training distribution. While fine-tuning can improve summarization quality, it typically relies on costly and scarce high-quality labeled data. In this work, we explore continual pre-training as a scalable, self-supervised approach to adapt LLMs for downstream summarization tasks, particularly in the context of noisy real-world conversation transcripts. We conduct extensive experiments using large-scale, unlabeled business conversation data to investigate whether continual pre-training enhances model capabilities in conversational summarization. Our results demonstrate that continual pre-training yields substantial gains in both in-domain and out-of-domain summarization benchmarks, while maintaining strong generalization and robustness. We also analyze the effects of data selection strategies, providing practical guidelines for applying continual pre-training in summarization-focused industrial applications.

---

### How to model Human Actions distribution with Event Sequence Data

Authors: Egor Surkov, Dmitry Osin, Evgeny Burnaev, Egor Shvetsov
Published: 2025-10-07
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.05856v1

This paper studies forecasting of the future distribution of events in human action sequences, a task essential in domains like retail, finance, healthcare, and recommendation systems where the precise temporal order is often less critical than the set of outcomes. We challenge the dominant autoregressive paradigm and investigate whether explicitly modeling the future distribution or order-invariant multi-token approaches outperform order-preserving methods. We analyze local order invariance and introduce a KL-based metric to quantify temporal drift. We find that a simple explicit distribution forecasting objective consistently surpasses complex implicit baselines. We further demonstrate that mode collapse of predicted categories is primarily driven by distributional imbalance. This work provides a principled framework for selecting modeling strategies and offers practical guidance for building more accurate and robust forecasting systems.

---

### ESS-Flow: Training-free guidance of flow-based models as inference in source space

Authors: Adhithyan Kalaivanan, Zheng Zhao, Jens Sjölund, Fredrik Lindsten
Published: 2025-10-07
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2510.05849v1

Guiding pretrained flow-based generative models for conditional generation or to produce samples with desired target properties enables solving diverse tasks without retraining on paired data. We present ESS-Flow, a gradient-free method that leverages the typically Gaussian prior of the source distribution in flow-based models to perform Bayesian inference directly in the source space using Elliptical Slice Sampling. ESS-Flow only requires forward passes through the generative model and observation process, no gradient or Jacobian computations, and is applicable even when gradients are unreliable or unavailable, such as with simulation-based observations or quantization in the generation or observation process. We demonstrate its effectiveness on designing materials with desired target properties and predicting protein structures from sparse inter-residue distance measurements.

---

### Luth: Efficient French Specialization for Small Language Models and Cross-Lingual Transfer

Authors: Maxence Lasbordes, Sinoué Gad
Published: 2025-10-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.05846v2

The landscape of Large Language Models remains predominantly English-centric, resulting in a significant performance gap for other major languages, such as French, especially in the context of Small Language Models (SLMs). Existing multilingual models demonstrate considerably lower performance in French compared to English, and research on efficient adaptation methods for French remains limited. To address this, we introduce \textbf{Luth}, a family of French-specialized SLMs: through targeted post-training on curated, high-quality French data, our models outperform all open-source counterparts of comparable size on multiple French benchmarks while retaining their original English capabilities. We further show that strategic model merging enhances performance in both languages, establishing Luth as a new state of the art for French SLMs and a robust baseline for future French-language research.

---

### Multimodal Trajectory Representation Learning for Travel Time Estimation

Authors: Zhi Liu, Xuyuan Hu, Xiao Han, Zhehao Dai, Zhaolin Deng, Guojiang Shen, Xiangjie Kong
Published: 2025-10-07
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.05840v2

Accurate travel time estimation (TTE) plays a crucial role in intelligent transportation systems. However, it remains challenging due to heterogeneous data sources and complex traffic dynamics. Moreover, traditional approaches typically convert trajectory data into fixed-length representations. This overlooks the inherent variability of real-world motion patterns, often resulting in information loss and redundancy. To address these challenges, this paper introduces the Multimodal Dynamic Trajectory Integration (MDTI) framework--a novel multimodal trajectory representation learning approach that integrates GPS sequences, grid trajectories, and road network constraints to enhance the performance of TTE. MDTI employs modality-specific encoders and a multimodal fusion module to capture complementary spatial, temporal, and topological semantics, while a dynamic trajectory modeling mechanism adaptively regulates information density for trajectories of varying lengths. Two self-supervised pretraining objectives, named contrastive alignment and masked language modeling, further strengthen multimodal consistency and contextual understanding. Extensive experiments on three real-world datasets demonstrate that MDTI consistently outperforms state-of-the-art baselines, confirming its robustness and strong generalization abilities. The code is publicly available at: https://github.com/City-Computing/MDTI.

---

### Towards Robust and Realible Multimodal Misinformation Recognition with Incomplete Modality

Authors: Hengyang Zhou, Yiwei Wei, Jian Yang, Zhenyu Zhang
Published: 2025-10-07
Categories: cs.MM, cs.CV
arXiv: https://arxiv.org/abs/2510.05839v3

Multimodal Misinformation Recognition has become an urgent task with the emergence of huge multimodal fake content on social media platforms. Previous studies mainly focus on complex feature extraction and fusion to learn discriminative information from multimodal content. However, in real-world applications, multimedia news may naturally lose some information during dissemination, resulting in modality incompleteness, which is detrimental to the generalization and robustness of existing models. To this end, we propose a novel generic and robust multimodal fusion strategy, termed Multi-expert Modality-incomplete Learning Network (MMLNet), which is simple yet effective. It consists of three key steps: (1) Multi-Expert Collaborative Reasoning to compensate for missing modalities by dynamically leveraging complementary information through multiple experts. (2) Incomplete Modality Adapters compensates for the missing information by leveraging the new feature distribution. (3) Modality Missing Learning leveraging an label-aware adaptive weighting strategy to learn a robust representation with contrastive learning. We evaluate MMLNet on three real-world benchmarks across two languages, demonstrating superior performance compared to state-of-the-art methods while maintaining relative simplicity. By ensuring the accuracy of misinformation recognition in incomplete modality scenarios caused by information propagation, MMLNet effectively curbs the spread of malicious misinformation. Code is publicly available at https://github.com/zhyhome/MMLNet.

---

### EEPO: Exploration-Enhanced Policy Optimization via Sample-Then-Forget

Authors: Liang Chen, Xueting Han, Qizhou Wang, Bo Han, Jing Bai, Hinrich Schutze, Kam-Fai Wong
Published: 2025-10-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.05837v2

Balancing exploration and exploitation remains a central challenge in reinforcement learning with verifiable rewards (RLVR) for large language models (LLMs). Current RLVR methods often overemphasize exploitation, leading to entropy collapse, diminished exploratory capacity, and ultimately limited performance gains. Although techniques that increase policy stochasticity can promote exploration, they frequently fail to escape dominant behavioral modes. This creates a self-reinforcing loop -- repeatedly sampling and rewarding dominant modes -- that further erodes exploration. We introduce Exploration-Enhanced Policy Optimization (EEPO), a framework that promotes exploration via two-stage rollouts with adaptive unlearning. In the first stage, the model generates half of the trajectories; it then undergoes a lightweight unlearning step to temporarily suppress these sampled responses, forcing the second stage to explore different regions of the output space. This sample-then-forget mechanism disrupts the self-reinforcing loop and promotes wider exploration during rollouts. Across five reasoning benchmarks, EEPO outperforms GRPO, achieving average relative gains of 24.3% on Qwen2.5-3B, 33.0% on Llama3.2-3B-Instruct, and 10.4% on Qwen3-8B-Base.

---

### Flow4Agent: Long-form Video Understanding via Motion Prior from Optical Flow

Authors: Ruyang Liu, Shangkun Sun, Haoran Tang, Ge Li, Wei Gao
Published: 2025-10-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.05836v1

Long-form video understanding has always been a challenging problem due to the significant redundancy in both temporal and spatial contents. This challenge is further exacerbated by the limited context length of Multimodal Large Language Models (MLLMs). To address this issue, many previous works have attempted to extract key video information, where the "key" is typically semantic-aware and heavily dependent on the CLIP model as prior. In this paper, we propose Flow4Agent, a novel framework that pioneeringly incorporates motion priors from optical flow to facilitate LLM-based long video understanding. Flow4Agent mitigates the redundancy in long videos at both temporal and spatial levels through two core modules: Temporal Granularity Optimization (TGO) adaptively refines framelevel hierarchies, which first leverages coarse flow priors to group similar visual contents and then applies semantic priors to filter out highly irrelevant scene information. Motion Token Pruning (MTP) further refines the intra-frame visual representations, pruning high-redundancy video tokens using fine-grained optical flow information. Extensive experiments demonstrate that our Flow4Agent outperforms existing methods across a wide range of video MLLM benchmarks, especially for hour-level video understanding tasks, achieving 64.7% on Video-MME, 71.4% on MLVU and 60.4% on LongVideoBench.

---

### FoleyGRAM: Video-to-Audio Generation with GRAM-Aligned Multimodal Encoders

Authors: Riccardo Fosco Gramaccioni, Christian Marinoni, Eleonora Grassucci, Giordano Cicchetti, Aurelio Uncini, Danilo Comminiello
Published: 2025-10-07
Categories: cs.SD, cs.CV, cs.LG, cs.MM, eess.AS
arXiv: https://arxiv.org/abs/2510.05829v1

In this work, we present FoleyGRAM, a novel approach to video-to-audio generation that emphasizes semantic conditioning through the use of aligned multimodal encoders. Building on prior advancements in video-to-audio generation, FoleyGRAM leverages the Gramian Representation Alignment Measure (GRAM) to align embeddings across video, text, and audio modalities, enabling precise semantic control over the audio generation process. The core of FoleyGRAM is a diffusion-based audio synthesis model conditioned on GRAM-aligned embeddings and waveform envelopes, ensuring both semantic richness and temporal alignment with the corresponding input video. We evaluate FoleyGRAM on the Greatest Hits dataset, a standard benchmark for video-to-audio models. Our experiments demonstrate that aligning multimodal encoders using GRAM enhances the system's ability to semantically align generated audio with video content, advancing the state of the art in video-to-audio synthesis.

---

### StereoSync: Spatially-Aware Stereo Audio Generation from Video

Authors: Christian Marinoni, Riccardo Fosco Gramaccioni, Kazuki Shimada, Takashi Shibuya, Yuki Mitsufuji, Danilo Comminiello
Published: 2025-10-07
Categories: cs.SD, cs.CV, cs.LG, cs.MM, eess.AS
arXiv: https://arxiv.org/abs/2510.05828v1

Although audio generation has been widely studied over recent years, video-aligned audio generation still remains a relatively unexplored frontier. To address this gap, we introduce StereoSync, a novel and efficient model designed to generate audio that is both temporally synchronized with a reference video and spatially aligned with its visual context. Moreover, StereoSync also achieves efficiency by leveraging pretrained foundation models, reducing the need for extensive training while maintaining high-quality synthesis. Unlike existing methods that primarily focus on temporal synchronization, StereoSync introduces a significant advancement by incorporating spatial awareness into video-aligned audio generation. Indeed, given an input video, our approach extracts spatial cues from depth maps and bounding boxes, using them as cross-attention conditioning in a diffusion-based audio generation model. Such an approach allows StereoSync to go beyond simple synchronization, producing stereo audio that dynamically adapts to the spatial structure and movement of a video scene. We evaluate StereoSync on Walking The Maps, a curated dataset comprising videos from video games that feature animated characters walking through diverse environments. Experimental results demonstrate the ability of StereoSync to achieve both temporal and spatial alignment, advancing the state of the art in video-to-audio generation and resulting in a significantly more immersive and realistic audio experience.

---