Reproduce “Uncertainty-Driven Replay Memory 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
Sheeraja Rajakrishnan, Alexander G. Ororbia, Travis Desell, Daniel E. Krutz
UDRM changes replay by retaining every transition and adding a second copy when epistemic uncertainty exceeds a threshold and reward is positive. It uses an evidential deep-learning setup. In Seaquest after 2.5 million steps, UDRM with recalibration every 50 steps reports average cumulative reward of 28.2662 ± 3.1341, versus 19.9233 ± 4.6946 for CEQR-DQN; the comparison does not isolate duplication’s effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.