Reproduce “Energy-Based Transfer for 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

Zeyun Deng, Jasorsi Ghosh, Fiona Xie, Kai Cheng, Yuzhe Lu, Katia Sycara, Joseph Campbell

Energy-Based Transfer Learning (EBTL) filters teacher advice by scoring whether a learner’s state resembles states familiar to the teacher, then advises only above a threshold when the schedule permits. In Alternating Goal MiniGrid, with a 96% optimal teacher, EBTL reports 46.1±4.5% relative transfer improvement over training from scratch, versus 6.1±5.6% for unfiltered advising, across 10 seeds; guidance ends halfway through training.

Summary by alphaXiv.

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