Reproduce “Vector Quantized Models for Planning” 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
Sherjil Ozair, Yazhe Li, Ali Razavi, Ioannis Antonoglou, Aäron van den Oord, Oriol Vinyals
VQM adds learned representations of possible environment responses to model-based tree search, allowing the planner to branch over outcomes as well as actions. It encodes outcomes as discrete latent codes and uses them in Monte Carlo tree search. In offline chess, VQ agents with worst-case chance nodes recover to two-player MuZeroChess performance; this result is specific to the evaluated setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.