About Me
I am Hanyu Wang, a third-year PhD student at Penn State University, fortunate to be advised by Prof. Jinghui Chen. Previously, I received my Bachelor of Science degree from the Department of Mathematics at University of Science and Technology of China (USTC) in June 2024.
My research lies at the intersection of Natural Language Processing and Trustworthy AI. I am broadly interested in a range of topics aimed at making AI systems more capable and reliable, including:
- Understanding and mitigating hallucinations in LLMs.
- Enhancing the multi-step reasoning abilities of language models.
- Building advanced agentic systems capable of reflection.
News
- 2026/10 Our new paper, “Balancing Reference Guidance and Free Generation in Trajectory Rollouts for Reasoning RL”, a result of my internship at Honda Research Institute, is now available on arXiv.
- 2026/10 Our new benchmark, “DecisionRobust: Benchmarking Robustness for JEV Style Decision Models Across Text and Vision”, is now available on arXiv, with a project page and code.
- 2026/09 Our paper, “ProD: Progress-Aware Distillation for Mitigating Stagnation in Small Language Model Agents”, was accepted to NeurIPS 2026! See you in Atlanta!
- 2026/09 Our paper, “Skill Transfer: Adapting Agent Skills from Strong to Weak Agents”, was accepted to the NeurIPS 2026 TTCL Workshop.
- 2026/05 Our new paper, “SkillGrad: Optimizing Agent Skills Like Gradient Descent”, is now available on arXiv, with code released on GitHub.
- 2026/05 Started my summer research internship at Honda Research Institute.
- 2026/05 Honored to be recognized as a Silver Reviewer at ICML 2026.
- 2026/02 Our paper, “PreFlect: From Retrospective to Prospective Reflection in Large Language Model Agents”, is now available on arXiv.
- 2025/05 Our paper, “TruthFlow: Truthful LLM Generation via Representation Flow Correction”, was accepted to ICML 2025!
- 2024/07 Our paper, “On the Discrepancy and Connection between Memorization and Generation in Diffusion Models”, was accepted to the ICML 2024 Workshop on Foundation Models in the Wild.
Selected Publications & Preprints
NeurIPS 2026 
Advances in Neural Information Processing Systems (NeurIPS 2026)
Preprint
ARGAdaptive Reference Guidance
ICML 2025 
Proceedings of the 42nd International Conference on Machine Learning (ICML 2025)
Academic Service
- Conference Reviewer: NeurIPS 2026, ICML 2026, ACL Rolling Review (January 2026, May 2025)
- Journal Reviewer: Transactions on Machine Learning Research (TMLR)
Contact
Email: hbw5365 at psu dot edu