The paper examines how large language models (LLMs) can be biased by irrelevant social contexts when evaluating teachers, using a large U.S. classroom transcript dataset. It shows that spurious contexts can shift model ratings by up to 1.48 points on a 7‑point scale and that standard mitigation methods like SFT and DPO are insufficient. The authors introduce Debiasing‑DPO, a method that combines contrastive reasoning‑augmented DPO with SFT, which reduces bias by 84% and improves predictive accuracy by 52% on Llama and Qwen Instruct models.
By Hyunji Nam, Dorottya Demszky
GRAPE is a two‑stage Bayesian optimization framework that first refines the local gradient posterior using a closed‑form acquisition function and then selects update directions by maximizing expected decrease conditioned on descent. The authors prove that the refinement stage monotonically reduces local uncertainty and that the progress‑aware direction converges to true steepest descent as the posterior sharpens. Empirical results show GRAPE achieves a 5.4× speedup on black‑box adversarial attacks and reduces final average regret by 3.8 log‑units on large language model prompt‑optimization tasks.
By Richard Cornelius Suwandi, Feng Yin
The paper investigates how explicit reasoning in Large Reasoning Models (LRMs) affects their ability to persuade and be persuaded. Experiments on objective and subjective tasks reveal a Persuasion Duality: reasoning boosts an agent’s persuasive power by about 21 percentage points while also making it less susceptible to incorrect persuasion by up to 10 percentage points. However, the study finds that persuasiveness often relies on superficial cues like response length and repetition rather than logical validity, and that persuasion can amplify or attenuate non‑linearly across multi‑hop agent chains. The authors also propose an attention‑guided prompt‑level adversarial argument detection method that improves agent robustness.
By Haodong Zhao, Jidong Li, Zhaomin Wu, Tianjie Ju, Zhuosheng Zhang, Bingsheng He, Gongshen Liu
The paper introduces COMLLM, a generative framework that combines Group Relative Policy Optimization with a Look‑Ahead Collaborative Simulation to enable multi‑turn reasoning for task offloading in Mobile Edge Computing. By performing multi‑step Monte Carlo rollouts that jointly model server queue dynamics, COMLLM incorporates long‑term system evolution into its reward design, achieving near‑optimal latency and improved load‑balancing fairness. The framework demonstrates zero‑shot scalability to larger network topologies, outperforming supervised fine‑tuning, deep reinforcement learning, and heuristic baselines without requiring retraining.
By Ning Yang, Chuangxin Cheng, Haijun Zhang
The paper introduces StablePDENet, a physics-informed adversarial training framework that regularizes the residual sensitivity of neural operators to improve stability against input perturbations. The method formulates operator learning as a min–max optimization, where a physics-based projected‑gradient adversary generates perturbations and the outer objective combines the attacked physics loss with a normalized residual‑sensitivity penalty. Experiments on benchmark problems show that StablePDENet outperforms PI‑DeepONet and its adversarial variant in accuracy under adversarial attacks while maintaining competitive performance on clean data, and it also enhances generalization and highlights the distinction between model sensitivity and intrinsic operator ill‑conditioning.
By Chutian Huang, Chang Ma, Kaibo Wang, Yang Xiang
The paper introduces CuCu, a multi‑agent LLM framework that converts national social studies curricula into open‑ended, culture‑specific question‑answer pairs for fine‑tuning language models. Using the Korean curriculum, the authors build KCaQA, a dataset of 34.1k QA pairs that cover culture‑specific topics and ground responses in local sociocultural contexts. Experiments show that fine‑tuning with KCaQA improves the model’s cultural alignment and relevance to Korean society.
By Haneul Yoo, Won Ik Cho, Geunhye Kim, Jiyoon Han
The paper evaluates privacy risks in graph neural networks enhanced by large language models (LLMs). Using a five‑stage framework, the authors test six real‑world text‑attributed graph datasets with 42 model configurations and six privacy attack methods across link, label, and membership inference threats. Results show that LLM‑enhanced GNNs are more vulnerable than shallow baselines, with semantic enrichment amplifying exploitable signals, and that differential privacy can reduce risk but at a significant cost to utility.
By Longzhu He, Zelang Wen, Chaozhuo Li, Sen Su
The paper introduces FairTPT, a fairness-aware test‑time prompt tuning method for vision‑language models like CLIP. It jointly minimizes target marginal entropy while maximizing spurious marginal entropy to reduce bias under subpopulation shifts. Experiments show that standard episodic test‑time adaptation can worsen disparities, but FairTPT outperforms existing debiasing methods while preserving overall performance.
By Yoann Launay, Parameswaran Kamalaruban, Tom Kempton, Stuart Burrell, David Sutton
The paper investigates hierarchical clustering under an individual fairness constraint that limits relative distortion within local k‑nearest neighborhoods. It formulates this as a feasibility problem over dominated ultrametrics, characterizes the minimal multiplicative slack needed, identifies a sharp local threshold, proves stability under bounded perturbations, establishes monotonicity in k, and demonstrates a Θ(log n) separation between local and global realizability. Experiments on synthetic and real‑world datasets corroborate the theoretical findings.
By Binita Maity, Shrutimoy Das
The paper investigates how refusal training shapes the internal geometry of language models, showing that activation updates from refusal-completion losses create a distinct low‑dimensional refusal subspace. In a case study on OLMo‑2‑0425‑1B‑Instruct, the authors link the brittleness of refusal directions to repetitive refusal prefixes and demonstrate that using diverse refusal starts can increase the stable rank of gradients, thereby hardening the model against vector‑ablation attacks. The work provides insights into the emergence of safety‑critical features and offers a potential strategy to strengthen refusal robustness.
By Andrey Labunets
The paper introduces a retrieval‑augmented multi‑agent framework that automatically generates instance‑specific evaluation rubrics for medical language models. By retrieving authoritative medical evidence, decomposing it into atomic facts, and combining these with user interaction constraints, the system produces fine‑grained criteria that outperform GPT‑4o on HealthBench and LLMEval‑Med. The generated rubrics also guide response refinement, improving medical LLM output quality by 9.2%.
By Yinzhu Chen, Abdine Maiga, Hossein A. Rahmani, Emine Yilmaz
VGA‑BenchV2 is an expanded, human‑aligned benchmark and optimization framework that jointly evaluates video generation quality and aesthetic value. It builds on the original VGA‑Bench taxonomy, adding 52 sub‑dimensions and 1,016 curated prompts to generate over 60,000 videos from 12 mainstream models. The benchmark significantly enlarges human supervision with 36,000 task‑level annotations and introduces a hybrid evaluator (VAQA‑Net, VTag‑Net, VGQA‑Net) that aligns well with human judgments and can be used as a reward model for reinforcement‑learning fine‑tuning.
By Longteng Jiang, DanDan Zheng, Qianqian Qiao, Heng Huang, Huaye Wang, Yihang Bo, Bao Peng, Jingdong Chen, Jun Zhou, Xin Jin
The paper introduces PuRF, a cache‑based method for multi‑label test‑time adaptation of vision‑language models. PuRF first performs region purification to reliably identify object‑specific regions, then uses these purified regions to build a discriminative, temporally refreshed cache that mitigates dominant‑label bias. Experiments show that PuRF outperforms existing methods, achieving a 4.05% mAP gain on ViT‑B/32 across five datasets.
By Yiwen Liang, Hui Chen, Yizhe Xiong, Mengyao Lyu, Yuhan Cao, Zijia Lin, Shuaicheng Niu, Sicheng Zhao, Jungong Han, Guiguang Ding
The paper presents a lightweight monocular vision pipeline for extracting sidewalk paths on low‑power embedded micromobility platforms. It evolves through three design iterations—from a skeleton‑graph baseline to a distance‑transform corridor planner and finally to a compact image‑space architecture—using a SegFormer‑B0 student model trained with semi‑supervised pseudo‑labels. The final system achieves high segmentation accuracy (IoU 0.946) and fast planning (under 50 ms per frame) while reducing temporal instability and increasing template‑path availability across real campus sequences.
By Lkhanaajav Mijiddorj, Yang Yan, Tyler Beringer, Bilguunzaya Mijiddorj, Alex N. Ho, Bin Xu, Binbin Weng
MyoMechanix is a multimodal dataset and framework for action quality assessment that incorporates muscle activity and other physiological signals alongside visual data. It contains over 7,500 samples of 20 weight‑loaded actions from 38 subjects, with synchronized RGB video, 3D pose, sEMG, and additional signals. The accompanying Fitness Knowledge Graph structures expert annotations into relationships among actions, phases, key steps, errors, and corrective feedback, enabling compositional scoring and interpretable assessment through the CUBIST engine. The project also introduces MyoMechanix‑AQA, MyoMechanix‑VideoQA, and a novel MyoMechanix‑Video2EMG task, demonstrating that multimodal sensing and structured representations improve performance, interpretability, and error attribution.
By Hao Yin, Paritosh Parmar, Lijun Gu, Lin Xu, Tianxiao Guo, Xiujin Liu, Tianyou Zheng, Yang Zhang, Weiwei Fu
The paper introduces PLUS, a framework that uses reinforcement learning to generate text-based summaries of individual users’ preferences, characteristics, and past conversations. These summaries condition a reward model, allowing it to predict personalized response preferences and improving reward accuracy by 11–77 % over the standard Bradley‑Terry model. PLUS demonstrates robust performance with new users and topics, achieves a 25 % improvement over existing personalized RLHF techniques, and enables zero‑shot personalization for state‑of‑the‑art models like GPT‑4.
By Hyunji Nam, Yanming Wan, Mickel Liu, Peter Ahnn, Jianxun Lian, Natasha Jaques
The paper investigates how different estimators of the reverse Kullback–Leibler (KL) divergence used as a regularization term in reinforcement learning (RL) training of large language models (LLMs) affect training stability and downstream performance. By analyzing gradient bias across various estimator configurations, the authors demonstrate that biased gradients can cause training instabilities, while unbiased configurations improve performance on both in‑domain and out‑of‑domain tasks. Experiments on Qwen2.5‑7B, Llama‑3.1‑8B‑Instruct, and Qwen3‑4B‑Instruct‑2507 confirm these findings and show that KL regularization also stabilizes off‑policy RL training in asynchronous setups.
By Vedant Shah, Johan Obando-Ceron, Vineet Jain, Brian Bartoldson, Bhavya Kailkhura, Sarthak Mittal, Glen Berseth, Pablo Samuel Castro, Yoshua Bengio, Esmeralda S. Whitammer, Moksh Jain, Siddarth Venkatraman, Aaron Courville
The paper introduces a diagnostic tool that measures how neural emulators of partial differential equations capture physical symmetries by evaluating the overlap of loss gradients along symmetry-related states. This metric probes the local geometry of the learned loss landscape and goes beyond traditional equivariance tests by directly assessing learning dynamics. Applied to autoregressive fluid flow emulators, the study shows that orbit-wise gradient coherence enables generalization over symmetry transformations and reveals when training selects a symmetry-compatible basin.
By James Amarel, Robyn Miller, Nicolas Hengartner, Benjamin Migliori, Emily Casleton, Alexei Skurikhin, Earl Lawrence, Gerd J. Kunde
SMART is a new framework that jointly tackles continuous sign language recognition (CSLR) and spotting by leveraging a multimodal large language model (MLLM) to generate motion descriptions as auxiliary semantic cues. It performs stable video‑text alignment with small batch sizes and introduces a Multi‑Scale Temporal Adapter to capture temporal interactions during transformer encoding. The framework also incorporates CSFormer, a CSLR‑guided spotting module that injects recognition‑derived gloss evidence into a boundary‑aware spotting network, enabling mutual benefit between recognition and spotting tasks.
By Eunjee Choi, JungHoon Sung, Seongwhan Cho, Chu Xin, Younggeun Choi
The paper introduces DeltaGateNet, a framework that models bidirectional temporal dynamics in EEG signals for driving fatigue recognition. It uses a Bidirectional Delta module to separate positive and negative temporal differences, and a Gated Temporal Convolution module to capture long‑term dependencies while preserving channel specificity. Experiments on SEED‑VIG and SADT datasets show that DeltaGateNet outperforms existing methods, achieving high intra‑subject and inter‑subject accuracies across balanced and unbalanced data.
By Yip Tin Po, Jianming Wang, Yutao Miao, Jiayan Zhang, Yunxu Zhao, Xiaomin Ouyang, Zhihong Li, Nevin L. Zhang