Reproduce “Free Random Projection for In-Context 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

Tomohiro Hayase, Benoît Collins, Nakamasa Inoue

The authors introduce free group random projections (FRP) for in-context reinforcement learning across environments. Each task’s input mapping is a product of random rotations, with mappings sharing word prefixes also sharing initial factors. The policy architecture remains unchanged. On held-out Stateless Cartpole, FRP(GRU) reaches ICL-Test MMER 4.08 ± 0.62, versus 2.90 ± 0.42 for RP(GRU), across 10 seeds, with no test-time weight updates.

Summary by alphaXiv.

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