Reproduce “Bayesian Distributional Policy Gradients” 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
Luchen Li, A. Aldo Faisal
BDPG models future-return distributions conditioned on states and uses posterior information gain to guide exploration. It also proves that its distributional Bellman backup contracts under a fixed policy. In Atari tests, BDPG without curiosity matched or outperformed distributional baselines in 6 of 8 games; curiosity-enabled BDPG scored highest among tested methods in the four hard-exploration games.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.