Publications

Full record on Google Scholar

Preprint 2026

Preprint

First author

From Discovery to Mastery: Reallocating the Replay Budget in Massively Parallel Reinforcement Learning

Guozheng Ma, Qi Zhao, Lu Li, Zilin Wang, Pierre-Luc Bacon, Dacheng Tao

Discovering successful behavior does not guarantee that a policy learns to reproduce it. REAP reallocates an existing replay budget toward high-return experience, improving learning across 48 control tasks without increasing interaction or update budgets.

Area
  • Core ML
  • Embodied AI
Themes
  • Scalable RL
  • Efficiency

ICML 2026

Published

Co-author

UltraHorizon: Benchmarking Agent Capabilities in Ultra Long-Horizon Scenarios

Haotian Luo*, Huaisong Zhang*, Xuelin Zhang*, Haoyu Wang*, Zeyu Qin*, Wenjie Lu*, Guozheng Ma, Haiying He, Yingsha Xie, Qiyang Zhou, Zixuan Hu, Hongze Mi, Yibo Wang, Naiqiang Tan, Hong Chen, Yi R. Fung, Chun Yuan, Li Shen * Equal contribution

A benchmark for sustained exploration, reasoning, memory and tool use over long trajectories in partially observable environments.

Area
  • LLMs
Themes
  • Exploration

arXiv 2026

Preprint

Co-author

Beyond Isolation: Unlocking Reinforcement Learning Component Synergy for Sample-Efficient Continuous Control

Qi Zhao, Guozheng Ma, Yilun Kong, Lu Li, Haoyu Wang, Zilin Wang, Tiantian Zhang, Yuxing Wang, Jian Sha, Yongzhe Chang, Xueqian Wang, Dacheng Tao

Coordinating representation learning, optimization stability and experience replay improves how continuous-control agents benefit from interacting algorithm components.

Area
  • Core ML
Themes
  • Efficiency