Reproduce “Submodular 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

Manish Prajapat, Mojmír Mutný, Melanie N. Zeilinger, Andreas Krause

Submodular Policy Optimization (SUBPO) is a policy-gradient method for reinforcement learning with history-dependent submodular rewards. It weights policy updates by the marginal gains that later states add after earlier visits, rather than summing each state’s standalone reward. In simulated tasks, its Markovian policy performs comparably to its history-conditioned version on gorilla-nest coverage, but this is not a general guarantee.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.