Hi, my name is Hongbin Lai. I am currently pursuing an M.S. at Peking University. My research interests lie in LLM Agent, Post-Training, and Data-Centric AI. As I continue to dive deeper into this evolving field, I am eager to explore innovative solutions, and I would sincerely appreciate guidance, discussions, and opportunities for collaboration.

I have published papers at top conferences. Google Scholar: citations.

🔥 News

• Oct 2026:  🍀 Our paper When Silence Becomes Distance: Diagnosing and Repairing Recall Loss in a Full-Duplex Speech Model is submitted to the ACL ARR October 2026 cycle (Co-author).
• Aug 2026:  📄 Our paper MinerU.Chem is available on arXiv (Co-contributor).
• Jul 2026:  🎉 Our paper HPFA is accepted by COLM 2026 (Co-first Author).
• Jul 2026:  🍀 Our paper StackPlanner is submitted to AAAI 2027, under review (Co-author).
• Mar 2026:  🎉 Our paper MolRecBench-Wild is accepted by CVPR 2026 Findings (Co-first Author).
• May 2025:  🎉 Our paper on Efficient Graph Continual Learning is accepted by ICML 2025 (Co-first Author).

📄 Preprints

📝 Publications

COLM 2026

HPFA: Hypergraph-Based Paired Failure Attribution for LLM Reasoning

Co-first Author

Paper

CVPR 2026 Findings

MolRecBench-Wild: A Real-World Benchmark for Optical Chemical Structure Recognition

Co-first Author

Paper / Dataset (Hugging Face) / Code

ICML 2025

Efficient Graph Continual Learning via Lightweight Graph Neural Tangent Kernels-based Dataset Distillation

Co-first Author

Paper

📖 Educations

• Sep 2024 – Jul 2027, M.S. in Electronic Information, School of Software and Microelectronics, Peking University
• Sep 2020 – Jul 2024, B.S. in Software Engineering, School of Computer Science (National Model Software College), Beijing University of Posts and Telecommunications