PosIR employs length-controlled bucketing and span-based relevance to diagnose position bias across 10 languages, revealing that models favor early content, especially in long documents.

Ziyang Zeng, Dun Zhang, Yu Yan et al.

Proposed a PAC-Bayesian framework optimizing anisotropic Gaussian posteriors for tighter generalization bounds.

Xinping Yi, Gaojie Jin, Xiaowei Huang et al.

JudgeFlow uses block responsibility scores to optimize LLM workflows, improving efficiency and interpretability.

Zihan Ma, Zhikai Zhao, Chuanbo Hua et al.

Using Kummer's theorem and prime decomposition, constructed infinitely many factorial divisibility examples with logarithmic gaps.

Nat Sothanaphan

This paper proves that supervised fine-tuning (SFT) and reinforcement learning (RL) in post-training are inherently coupled; RL degrades SFT, and vice versa.

Xueyan Niu, Bo Bai, Wei Han et al.

Proposes the Semantic Lifecycle framework leveraging foundation models to enhance semantic acquisition, representation, and storage for embodied AI.

Shuai Chen, Hao Chen, Yuanchen Bei et al.

Introduced NoisyBench benchmark; state-of-the-art models drop up to 80% in noisy environments; RARE improves robustness.

Seongyun Lee, Yongrae Jo, Minju Seo et al.

PALM employs structured affordance prediction and progress estimation with diffusion models, achieving 91.8% success in long-horizon robotic tasks.

Yuanzhe Liu, Jingyuan Zhu, Yuchen Mo et al.

Dr. Zero achieves search capabilities comparable to supervised learning through self-evolution and HRPO without training data.

Zhenrui Yue, Kartikeya Upasani, Xianjun Yang et al.

Proposes a post-training atomic task decomposition and fusion algorithm, achieving 98.2% accuracy in reentrancy vulnerability detection with enhanced out-of-distribution generalization.

Ying Zhou, Jiacheng Wei, Yu Qi et al.

Introduces personality-aware RL with D3QN and P4G dataset to improve persuasive dialogue outcomes.

Donghuo Zeng, Roberto Legaspi, Kazushi Ikeda

SpatialNav uses spatial scene graphs for zero-shot VLN, reaching 57.7% SR on R2R and 64.0% on R2R-CE.

Jiwen Zhang, Zejun Li, Siyuan Wang et al.

Proposes hyperspherical contrastive learning to embed emotions in a circular structure, improving interpretability and psychological alignment.

Yusuke Yamauchi, Akiko Aizawa

Proposes a unified attribution pipeline with four modules to mitigate hallucinations in RAG systems, significantly improving response fidelity.

Yuqing Zhao, Ziyao Liu, Yongsen Zheng et al.

Mechanistic analysis of DiT reveals two distinct circuits for spatial relation generation: stepwise attention heads with random embeddings and integrated fusion with T5.

Binxu Wang, Jingxuan Fan, Xu Pan

TAGRPO boosts GRPO in image-to-video generation via direct trajectory alignment, achieving significant reward improvements.

Jin Wang, Jianxiang Lu, Guangzheng Xu et al.

Proposes Self-Evolving Distributed Memory Architecture with three-layer coordination, achieving 87.3% memory efficiency and 142.5 ops/sec.

Zixuan Li, Chuanzhen Wang, Haotian Sun

Introduces Task Cascades framework, combining operation, document portion, and model choices, reducing inference costs by 36%.

Shreya Shankar, Sepanta Zeighami, Aditya Parameswaran

ALVGL employs sparse+low-rank decomposition with ADMM to enhance causal discovery efficiency and accuracy in high-dimensional and confounded data.

Pingchuan Ma, Qixin Zhang, Shuai Wang et al.

DafnyPro enhances LLM-based verification annotation generation via diff-checking, pruning, and proof strategy retrieval, achieving 86% correctness on DafnyBench.

Debangshu Banerjee, Olivier Bouissou, Stefan Zetzsche