Reproduce “DeepAveragers: Offline Reinforcement Learning by Solving Derived Non-Parametric MDPs” 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
Aayam Shrestha, Stefan Lee, Prasad Tadepalli, Alan Fern
DeepAveragers turns a fixed log into a planning problem and finds a policy through value iteration. It represents states, uses nearest logged examples to estimate rewards and transitions, and penalizes actions poorly supported by the data. In a tested 3D navigation task, replanning around a disabled left turn achieved average reward 0.96, supporting adaptation in that setup, not arbitrary failures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.