Reproduce “Hyper: Hyperparameter Robust Efficient Exploration in Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yiran Wang, Chenshu Liu, Yunfan Li, Sanae Amani, Bolei Zhou, Lin F. Yang
The HYPER algorithm enhances curiosity-driven reinforcement learning by reducing sensitivity to the intrinsic reward coefficient
Key insights
- Decoupling exploitation and exploration policies prevents large intrinsic rewards from corrupting the task value function, enhancing optimization stability.
- Repositioning the agent to task-relevant states before curiosity-driven exploration regularizes state visitation, mitigating instability and preventing excessive, undirected exploration.
- A dynamically adjusted repositioning phase enables exploration to launch from progressively more advantageous starting points, improving the quality of data collected for the exploitation policy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.