News & Research

arXiv:2507.15336v3 Announce Type: replace-cross Abstract: Designing high-performance neural networks for new tasks requires balancing optimization quality with search efficiency. Current methods fail to achieve this balance: neural architectural search is computationally expensive, while model retrieval often yields suboptimal static checkpoints. To resolve this dilemma, we model the performance gains induced by fine-grained architectural modifications as edit-effect evidence and build evidence graphs from prior tasks. By constructing a retrieval-augmented model refinement framework, our proposed M-DESIGN dynamically weaves historical evidence to discover near-optimal modification paths. M-DESIGN features an adaptive retrieval mechanism that quickly calibrates the evolving transferability of edit-effect evidence from different sources. To handle out-of-distribution shifts, we introduce predictive task planners that extrapolate gains from multi-hop evidence, thereby reducing reliance on

arXiv:2507.19881v2 Announce Type: replace-cross Abstract: Federated domain generalization has shown promising progress in image classification by enabling collaborative training across multiple clients without sharing raw data. However, its potential in the semantic segmentation of autonomous driving remains underexplored. In this paper, we propose FedS2R, the first one-shot federated domain generalization framework for synthetic-to-real semantic segmentation in autonomous driving. FedS2R comprises two components: an inconsistency-driven data augmentation strategy that generates images for unstable classes, and a multi-client knowledge distillation scheme with feature fusion that distills a global model from multiple client models. Experiments on five real-world datasets, Cityscapes, BDD100K, Mapillary, IDD, and ACDC, show that the global model significantly outperforms individual client models and is only 2 mIoU points behind the model trained with simultaneous access to all client da

arXiv:2508.01815v2 Announce Type: replace-cross Abstract: Text-to-SPARQL maps natural-language questions to executable SPARQL queries over RDF knowledge graphs. While standard evaluations often fix the target graph in advance, practical knowledge graph question answering (KGQA) may involve heterogeneous graph collections with different schemas, partial alignments, and incomplete metadata. In this setting, query generation depends on more than SPARQL syntax: the system must identify a graph schema that can support the predicates, entity types, joins, filters, and constraints required by the question. We present SchemaForge, a schema-grounded agentic framework for text-to-SPARQL over heterogeneous KG collections. Its central mechanism is question-conditioned schema-slice alignment: weak graph evidence first identifies plausible graphs, while stronger schema evidence determines whether a local schema slice can realize the intended query. The selected schema slice then constrains query gen

arXiv:2508.08337v3 Announce Type: replace-cross Abstract: Algorithmic fairness research has largely framed unfairness as discrimination along sensitive attributes. However, this approach limits visibility into unfairness as structural injustice instantiated through social determinants, which are contextual variables that shape attributes and outcomes without pertaining to specific individuals. This position paper argues that the field should quantify structural injustice via social determinants, beyond sensitive attributes. Drawing on cross-disciplinary insights, we argue that prevailing technical paradigms fail to adequately capture unfairness as structural injustice, because contexts are potentially treated as noise to be normalized rather than signal to be audited. We further demonstrate the practical urgency of this shift through a theoretical model of college admissions, a demographic study using U.S. census data, and a high-stakes domain application regarding breast cancer screen

arXiv:2508.12551v2 Announce Type: replace-cross Abstract: Linux kernel tuning is essential for optimizing operating system (OS) performance, yet remains challenging due to the complex kernel space, sparse performance feedback, and strong workload sensitivity. We present TuneAgent, an agentic Linux kernel tuning framework powered by rule-based reinforcement learning (RL). TuneAgent formulates the kernel space as a constrained RL environment, enabling large language models (LLMs) to autonomously explore the kernel while enforcing valid and precise configuration modifications. To address sparse performance feedback, we design structured reward functions that jointly promote reasoning standardization, configuration correctness, and performance awareness. Furthermore, we propose a two-phase training strategy that first ensures format and semantic correctness and then transitions to performance-driven exploration, accelerating convergence and reducing overhead. Experimental results show that

arXiv:2509.05367v5 Announce Type: replace-cross Abstract: Large Language Model safety alignment predominantly operates on a binary assumption that requests are either safe or unsafe. This classification proves insufficient when models encounter ethical dilemmas, where the capacity to reason through moral trade-offs creates a distinct attack surface. We formalize this vulnerability through TRIAL, a multi-turn red-teaming methodology that embeds harmful requests within ethical framings. TRIAL achieves high attack success rates across most tested models by systematically exploiting the model's ethical reasoning capabilities to frame harmful actions as morally necessary compromises. Building on these insights, we introduce ERR (Ethical Reasoning Robustness), a defense framework that distinguishes between instrumental responses that enable harmful outcomes and explanatory responses that analyze ethical frameworks without endorsing harmful acts. ERR employs a Layer-Stratified Harm-Gated LoRA

arXiv:2509.06093v4 Announce Type: replace-cross Abstract: Materials synthesis procedures are predominantly documented as narrative text in papers, protocols, and laboratory records, placing them beyond the reach of conventional data-driven optimization frameworks. This language-native character poses a particular challenge for complex, multistage processes such as the preparation of boron nitride nanosheets (BNNS), where outcomes depend on path-dependent choices in exfoliation, functionalization, and functionalization. Here, we recast synthesis planning of the materials as a text reasoning problem enabled by a lightly structured knowledge substrate that preserves the procedural logic and causal contexts while exposing computable elements for retrieval. Built on this representation, our framework combines semantic matching, lexical search, and parameter-aware filtering to support retrieval-augmented generation with more accurate and better-grounded synthesis guidance. We further introdu

arXiv:2509.13805v4 Announce Type: replace-cross Abstract: Foundation models have revolutionized natural language processing through a ``train once, deploy anywhere'' paradigm, where a single pre-trained model adapts to countless downstream tasks without retraining. Access to a Physics Foundation Model (PFM) would be transformative - democratizing access to high-fidelity simulations, accelerating scientific discovery, and eliminating the need for specialized solver development. Yet current physics-aware machine learning approaches remain fundamentally limited to single, narrow domains and require retraining for each new system. We present the General Physics Transformer (GPhyT), trained on 1.8 TB of diverse simulation data, that demonstrates foundation model capabilities are achievable for physics. Our key insight is that transformers can learn to infer governing dynamics from context, enabling a single model to simulate fluid-solid interactions, shock waves, thermal convection, and mul

arXiv:2509.24696v2 Announce Type: replace-cross Abstract: Personalizing large language models (LLMs) to individual user preferences is a critical step beyond generating generically helpful responses. However, current personalization methods are ill-suited for new users, as they typically require either slow, resource-intensive fine-tuning or a substantial amount of pre-existing user data, creating a significant cold-start problem. To address this challenge, we introduce a new paradigm for real-time personalization by learning from online pairwise preference feedback collected during text generation. We propose T-POP (Test-Time Personalization with Online Preference Feedback}), a novel algorithm that synergistically combines test-time alignment with dueling bandits. Without updating the LLM parameters, T-POP steers the decoding process of a frozen LLM by learning a reward function online that captures user preferences. By leveraging dueling bandits, T-POP intelligently queries the user

arXiv:2509.25773v3 Announce Type: replace-cross Abstract: AI models capable of comprehending humor hold real-world promise -- for example, enhancing engagement in human-machine interactions. To gauge and diagnose the capacity of multimodal large language models (MLLMs) for humor understanding, we introduce v-HUB, a novel video humor understanding benchmark. v-HUB comprises a curated collection of non-verbal short videos, reflecting real-world scenarios where humor can be appreciated purely through visual cues. We pair each video clip with rich annotations to support a variety of evaluation tasks and analyses, including a novel study of environmental sound that can enhance humor. To broaden its applicability, we construct an open-ended QA task, making v-HUB readily integrable into existing video understanding task suites. We evaluate a diverse set of MLLMs, from specialized Video-LLMs to versatile OmniLLMs that can natively process audio, covering both open-source and proprietary domain

arXiv:2509.25837v3 Announce Type: replace-cross Abstract: Large language models (LLMs) deliver remarkable performance but are costly to deploy, motivating knowledge distillation (KD) for efficient inference. Existing KD objectives typically match student and teacher probabilities via softmax, which blurs valuable logit information. While direct logit distillation (DLD) mitigates softmax smoothing, it fails to account for logit shift invariance, thereby restricting the solution space. We propose Concrete Score Distillation (CSD), a discrete score-matching objective that overcomes both softmax-induced smoothing and restrictions on the optimal solution set. We resolve the training instability and quadratic complexity of discrete score-matching in autoregressive LLMs, and the resulting CSD objective aligns relative logit differences across all vocabulary pairs between student and teacher with flexible weighting. We provide both mode-seeking and mode-covering instances within our framework

arXiv:2510.00481v2 Announce Type: replace-cross Abstract: In 2025, Large Language Model (LLM) services have launched a new feature -- AI video chat -- allowing users to interact with AI agents via real-time video communication (RTC), just like chatting with real people. Despite its significance, no systematic study has characterized the performance of existing AI video chat systems. To address this gap, this paper proposes a comprehensive benchmark across four dimensions: quality, latency, internal mechanisms, and system overhead. Using custom testbeds, we further evaluate six mainstream AI video chatbots with this benchmark. We also build an online platform for user study. The measurement leads to interesting findings that could be beneficial to the future optimizations. For example, the network latency of AI video chat matters not as much as human video chat. The capabilities of AI agents matters most in the user experience. Our benchmarking results also open up several research ques

arXiv:2510.01167v2 Announce Type: replace-cross Abstract: Aligning large language models to human preferences is inherently multidimensional, yet most pipelines collapse heterogeneous signals into a single objective. We seek to answer what it would take to simultaneously align a model across various domains spanning those with: verifiable rewards, non-verifiable subjective preferences, and complex interactive scenarios. Such multi-objective alignment setups are often plagued by individual objectives being at odds with each other, resulting in inefficient training and limited user control during inference. To address these issues, we propose $\textbf{M}$ulti-$\textbf{A}$ction-$\textbf{H}$ead $\textbf{AL}$ignment with PRM-guided Dec$\textbf{O}$ding ($\textbf{MAHALO}$), a unified framework that standardizes PRM training across verifiable and non-verifiable settings for step-level supervision, performs vectorized multi-objective alignment with Multi-Action-Head DPO, and enables controllabl

arXiv:2510.03259v2 Announce Type: replace-cross Abstract: Recent research on reasoning models explores the meta-awareness of language models, including their ability to determine optimal thinking duration, recognize knowledge boundaries, and structure concept-level thinking. While current large reasoning models depend solely on answer-based verification, we show that adding meta-awareness objectives leads to significant performance gains over models without such meta-knowledge. MAPR (Meta-Awareness via Predictive Reward) utilizes a self-generated task of predicting rollout statistics - specifically length, pass-rate, and concepts used - allowing for verification against the actual statistics. Furthermore, by leveraging this self-predictive capability, the model can regulate its reasoning behavior by i) filtering out trivial or unsolvable prompts, ii) reducing lengthy generations that tend to be incorrect, and iii) generating hints relevant to the problem. The results are inspiring: MAP

arXiv:2510.05342v2 Announce Type: replace-cross Abstract: Direct Preference Optimization (DPO) has emerged as a simple and effective method for aligning large language models. However, its reliance on a fixed temperature parameter leads to suboptimal training on diverse preference data, causing overfitting on easy examples and under-learning from informative ones. Recent methods have emerged to counter this. While IPO addresses general overfitting, its uniform regularization can be overly conservative. The more targeted approach of $\beta$-DPO suffers from its own limitations: its batch-level adaptation applies a single, compromised temperature to mixed-margin pairs, its linear update rule can produce unstable negative $\beta$ values, and its filtering mechanism discards potentially useful training signals. In this work, we introduce Margin-Adaptive Direct Preference Optimization (MADPO), a method that provides a stable, data-preserving, and instance-level solution. MADPO employs a pra