SCOUT framework decouples exploration and exploitation, enabling Qwen2.5-3B-Instruct to score 0.86 on unseen tasks, saving 60% GPU hours.

Haoyu Wang, Guozheng Ma, Shugang Cui et al.

KAPSO employs a knowledge-grounded long-horizon optimization framework integrating git isolation, structured knowledge graphs, and episodic memory to enhance autonomous program synthesis.

Alireza Nadafian, Alireza Mohammadshahi, Majid Yazdani

MAD method reduces cross-modal hallucinations by adaptive decoding, achieving 7.8% and 2.0% improvements on CMM and AVHBench.

Sangyun Chung, Se Yeon Kim, Youngchae Chee et al.

AMA framework achieves adaptive memory via multi-agent collaboration, reducing token consumption by 80%.

Weiquan Huang, Zixuan Wang, Hehai Lin et al.

PROTEUS uses Lagrangian RL for SLA-aware multi-LLM routing, achieving 94.0% accuracy and 89.8% cost savings.

Amit Singh Bhatti, Vishal Vaddina, Dagnachew Birru

Yunjue Agent achieves zero-start adaptability for open tasks via parallel batch evolution and tool optimization, outperforming baselines.

Haotian Li, Shijun Yang, Weizhen Qi et al.

Proposes DSGym, a modular framework for end-to-end evaluation and training of data science agents, covering diverse scientific and predictive tasks.

Fan Nie, Junlin Wang, Harper Hua et al.

EvoCUA employs a self-evolving cycle combining verifiable synthetic task generation, large-scale asynchronous interaction, and iterative policy optimization, achieving 56.7% success on OSWorld.

Taofeng Xue, Chong Peng, Mianqiu Huang et al.

PhysProver combines RLVR and PhysLeanData, enhancing physics theorem proving by 2.4%.

Hanning Zhang, Ruida Wang, Rui Pan et al.

ReCreate employs experience-driven, interaction-based scaffold updates to automate domain agent creation, outperforming black-box methods with 5%+ gains.

Zhezheng Hao, Hong Wang, Jian Luo et al.

LADFA combines LLMs and RAG to analyze personal data flows in privacy policies.

Haiyue Yuan, Nikolay Matyunin, Ali Raza et al.

EAPO enhances long-context reasoning with reward co-evolution, achieving significant performance gains.

Xin Guan, Zijian Li, Shen Huang et al.

PersonaDual balances personalization and objectivity via adaptive reasoning, improving accuracy by 3%.

Xiaoyou Liu, Xinyi Mou, Shengbin Yue et al.

Proposes TabGR using Attributed Table Graph and QG-PPR to improve table reasoning accuracy by 9.7%.

Yuxiang Wang, Junhao Gan, Shengxiang Gao et al.

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

Zihan Ma, Zhikai Zhao, Chuanbo Hua 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.

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.

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

SmartSearch enhances search agent query quality via process rewards and query refinement, significantly boosting retrieval efficiency.

Tongyu Wen, Guanting Dong, Zhicheng Dou

XGrammar-2 introduces TagDispatch and cross-grammar cache, achieving 6× faster structured generation for dynamic agent workloads.

Linzhang Li, Yixin Dong, Guanjie Wang et al.