InfiMed integrates reflective-patterned CoT and rule-based RLVR, boosting low-resource medical multimodal model performance, achieving 7 SOTA benchmarks.

Zeyu Liu, Zhitian Hou, Guanghao Zhu et al.

WorkForceAgent-R1 enhances LLM web agents' reasoning via R1-style RL, achieving 10.26-16.59% improvement.

Yuchen Zhuang, Di Jin, Jiaao Chen et al.

WebDancer, based on ReAct, achieves 64.1% Pass@1 on GAIA, significantly improving multi-step web reasoning.

Jialong Wu, Baixuan Li, Runnan Fang et al.

Combining orthogonality and variance losses enhances MoE expert specialization, achieving up to 23.79% performance gain.

Hongcan Guo, Haolang Lu, Guoshun Nan et al.

BioHopR is a biomedical multi-hop, multi-answer reasoning benchmark based on PrimeKG, evaluating large language models' inference capabilities.

Yunsoo Kim, Yusuf Abdulle, Honghan Wu

This study evaluates LLMs in medical summarization, emphasizing vocabulary adaptation's role in high OOV scenarios.

Gunjan Balde, Soumyadeep Roy, Mainack Mondal et al.

Proposed a logit-based ensemble method matching human evaluation of LLM consistency.

Xiaoyuan Wu, Weiran Lin, Omer Akgul et al.

Proposes a dynamic multi-agent orchestration framework using reinforcement learning, improving task performance by 15% and reducing tokens by 30%.

Yufan Dang, Chen Qian, Xueheng Luo et al.

Reinforcement fine-tuning enhances reasoning in multimodal LLMs, significantly boosting OpenAI-o1 and DeepSeek-R1 performance.

Haoyuan Sun, Jiaqi Wu, Bo Xia et al.

TriSense introduces a multimodal LLM with adaptive query-based fusion, trained on TriSense-2M with 2 million long videos, achieving state-of-the-art results in video understanding.

Zinuo Li, Xian Zhang, Yongxin Guo et al.

AlignXplore model infers personalized preferences from behavioral signals, achieving a 15.49% performance boost.

Jia-Nan Li, Jian Guan, Wei Wu et al.

QwenLong-L1 employs progressive context scaling and RL algorithms (GRPO, DAPO) to enhance long-text reasoning, outperforming existing models.

Fanqi Wan, Weizhou Shen, Shengyi Liao et al.

Proposes Agent Distillation, transferring LLM agent behaviors into small models with retrieval and code tools, achieving performance comparable to larger models.

Minki Kang, Jongwon Jeong, Seanie Lee et al.

ToDi adaptively combines FKL and RKL per token via probability ratio, significantly improving distillation accuracy.

Seongryong Jung, Suwan Yoon, DongGeon Kim et al.

IRONIC framework achieves state-of-the-art zero-shot sarcasm detection using multi-modal coherence relations.

Aashish Anantha Ramakrishnan, Aadarsh Anantha Ramakrishnan, Dongwon Lee

GUI-G1 employs improved RL strategies and reward design, achieving 90.3% accuracy on 17K samples, surpassing similar-sized models.

Yuqi Zhou, Sunhao Dai, Shuai Wang et al.

Soft Thinking generates continuous concept tokens, boosting accuracy by 2.48% and reducing tokens by 22.4% without training.

Zhen Zhang, Xuehai He, Weixiang Yan et al.

Hunyuan-TurboS combines Mamba-Transformer with adaptive chain-of-thought, 560B parameters, achieving top-tier performance.

Tencent Hunyuan Team, Ao Liu, Botong Zhou et al.

BanglaByT5 is the first byte-level model for Bangla, outperforming multilingual models.

Pramit Bhattacharyya, Arnab Bhattacharya

Utilizing PMP for explainable sarcasm detection in Australian and Indian English, achieving significant performance improvements.

Ishmanbir Singh, Dipankar Srirag, Aditya Joshi