Publications

2026

1. MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning

   Yixuan Xie, Xueqiang Xu , Xinyi Fan , Xinxi Lyu , Tao Feng , Xing Fan , Chenlei Guo , Jie Hao , Shaowen Wang , Jiashuo Sun , and Jiawei Han

   In submission to ARR Oct, 2026

   TL;DR arXiv Code Hugging Face

   TL;DR: Retrieving agent memories by similarity is not enough; we train a memory composer with reinforcement learning to jointly select and order an LLM agent’s past experiences, improving task success with fewer memory tokens.

2. CaseMem: Closing the Loop from Agent Outcomes to Memory Reuse

   Xinyi Fan , Yixuan Xie, Xing Fan , Chenlei Guo , Jie Hao , and Jiawei Han

   In submission to ICLR, 2027
3. Ask to Distill: Learning to Synthesize Data for On-Policy Distillation

   Jiashuo Sun , Hao Zhang , Xueqiang Xu , Heng Wang , Yixuan Xie, Pengcheng Jiang , Zhenrui Yue , Jiajun Fan , Peiran Li , Ge Liu , Shaowen Wang , and Jiawei Han

   Code Hugging Face
4. Retrieval is Cheap, Show Me the Code: Executable Multi-Hop Reasoning for Retrieval-Augmented Generation

   Jiashuo Sun^† , Jimeng Shi^† , Yixuan Xie, Saizhuo Wang , Jash Rajesh Parekh , Pengcheng Jiang , Zhiyi Shi , Jiajun Fan , Qinglong Zheng , Peiran Li , Shaowen Wang , Ge Liu , and Jiawei Han

   TL;DR arXiv Project Page Code

   TL;DR: PyRAG reformulates multi-hop RAG as executable Python programs over retrieve/answer tools, with compiler-grounded self-repair and adaptive retrieval—outperforming strong baselines on five QA benchmarks (up to +25.5 EM on Bamboogle).

5. TaSR-RAG: Taxonomy-guided Structured Reasoning for Retrieval-Augmented Generation

   Jiashuo Sun^† , Yixuan Xie^†, Jimeng Shi , Shaowen Wang , and Jiawei Han

   TL;DR arXiv

   TL;DR: We propose TaSR-RAG, which uses a lightweight two-level taxonomy to guide step-wise triple-based evidence selection for multi-hop RAG, outperforming strong baselines by up to 14%.