Reproduce “Grid Cells as Spectral Successor Features: A Reusable Basis for Continual 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
Élodie Côté-Gauthier, Doina Precup, Isabeau Prémont-Schwarz
The authors use spectral successor features to help reinforcement-learning agents adapt when goals change within a fixed environment. They represent state-action pairs with eigenvectors of a reference-policy transition operator, allowing successor features to be calculated and reward weights to be updated without a learned successor-feature network. In a deterministic gridworld, spectral agents retained near-maximum return across two-goal switches, but comparisons had unequal information.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.