Reproduce “Beyond Fine-Tuning: Transferring Behavior in 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

Víctor Campos, Pablo Sprechmann, Steven Hansen, Andre Barreto, Steven Kapturowski, Alex Vitvitskyi, Adrià Puigdomènech Badia, Charles Blundell

The authors introduce BT: a task-learning agent consults a fixed reward-free policy while learning from task rewards. The agent can use its behavior through policy flights or a selectable pseudo-action. After 5 billion frames, R2D2 with BT scored a median 193.15 across six hard-exploration Atari games, versus 103.62 for R2D2 with εz-greedy; these are scores, not measured time or compute savings.

Summary by alphaXiv.

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