MedVLM-R1 employs reinforcement learning with GRPO to generate explicit reasoning, boosting accuracy to 78.22% on medical VQA tasks.

Jiazhen Pan, Che Liu, Junde Wu et al.

Bi'an introduces a bilingual benchmark and lightweight models, achieving 88.2% accuracy in RAG hallucination detection, close to GPT-4o performance.

Zhouyu Jiang, Mengshu Sun, Zhiqiang Zhang et al.

CLADA framework enhances LLM efficiency with cognitive-load-aware dynamic activation, achieving 20% speedup with less than 2% accuracy drop.

Yiheng Yang, Yujie Wang, Chi Ma et al.

OneRec employs an end-to-end generative model with session-wise decoding and preference alignment, achieving 1.6% watch-time increase in industry deployment.

Jiaxin Deng, Shiyao Wang, Kuo Cai et al.

Multi-agent Gemini-based system accelerates hypothesis generation, improving Elo scores from 3.0 to 4.2 over iterative cycles, validated in drug repurposing and target discovery.

Juraj Gottweis, Wei-Hung Weng, Alexander Daryin et al.

ConsJudge method improves RAG model evaluation consistency, enhancing LLM judgment accuracy.

Shuliang Liu, Xinze Li, Zhenghao Liu et al.

Proposes Self-Certainty, a KL-divergence-based metric leveraging LLM output probabilities for scalable Best-of-N response selection without external reward models.

Zhewei Kang, Xuandong Zhao, Dawn Song

The paper introduces a nested kernel quadrature estimator to reduce sample requirements and improve convergence speed.

Zonghao Chen, Masha Naslidnyk, François-Xavier Briol

This paper systematically explores the design space for info extraction from layout-rich documents using LLMs, introducing LayIE-LLM and achieving up to 37.5 F1 improvement.

Gaye Colakoglu, Gürkan Solmaz, Jonathan Fürst

SpargeAttn introduces a training-free universal sparse attention method, accelerating inference by 2.5-5x across language, image, and video models without performance loss.

Jintao Zhang, Chendong Xiang, Haofeng Huang et al.

GazeBot integrates gaze information and motion bottlenecks with Transformer-based prediction to enhance skill generalization in robot manipulation, achieving success rates over 85% in unseen scenarios.

Ryo Takizawa, Izumi Karino, Koki Nakagawa et al.

ViDoRAG combines GMM-based multi-modal retrieval with multi-agent iterative reasoning, achieving over 10% improvement on ViDoSeek.

Qiuchen Wang, Ruixue Ding, Zehui Chen et al.

This survey frames video generation as sequential sampling from a high-dimensional spatiotemporal distribution and compares VAE, AR, diffusion, and flow models.

Zhiyu Yin, Kehai Chen, Xuefeng Bai et al.

Proposes multimodal LLMs for zero-shot time series anomaly detection; introduces VisualTimeAnomaly benchmark; finds coarse anomalies better detected than fine ones.

Xiongxiao Xu, Haoran Wang, Yueqing Liang et al.

Gradient-based method uncovers multi-dimensional refusal cones in LLMs, revealing independent mechanisms governing refusal behavior.

Tom Wollschläger, Jannes Elstner, Simon Geisler et al.

PosterSum employs a hierarchical segmentation and summarization approach, improving ROUGE-L by 3.14% for scientific poster abstracts.

Rohit Saxena, Pasquale Minervini, Frank Keller

VideoGrain modulates space-time attention for fine-grained video editing, achieving state-of-the-art results in zero-shot multi-grained control.

Xiangpeng Yang, Linchao Zhu, Hehe Fan et al.

AutoLogi generates logic puzzles with program verification and controllable difficulty, evaluating LLM reasoning from 35% to 73%.

Qin Zhu, Fei Huang, Runyu Peng et al.

ONPO uses optimistic mirror descent for general-preference alignment, achieving O(1/T) gap and AlpacaEval scores up to 48.6.

Yuheng Zhang, Dian Yu, Tao Ge et al.

CODESYNC enhances LLMs' code synchronization by updating Python library APIs in real-time.

Chenlong Wang, Zhaoyang Chu, Zhengxiang Cheng et al.