Reproduce “VariBAD: A Very Good Method for Bayes-Adaptive Deep RL via Meta-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
Luisa Zintgraf, Kyriacos Shiarlis, Maximilian Igl, Sebastian Schulze, Yarin Gal, Katja Hofmann, Shimon Whiteson
VariBAD learns a task-uncertainty representation from related reinforcement-learning tasks and gives it to a policy for action selection. A recurrent encoder infers a stochastic task embedding from interaction history, while a decoder trains it to predict states and rewards. In an unknown-goal gridworld, variBAD reaches privileged optimal performance by rollout three, close to Bayes-optimal, but this result is task-specific.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.