Reproduce “Diversity-Aware Experience Replay via Statistical Leverage Scores” 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
Shaifalee Saxena, Alexander Scheinker
The authors introduce leverage-based replay sampling, scoring unusual stored transitions from buffer-derived feature geometry and mixing those scores with uniform sampling. For relational behavior, they score 24-transition windows using goal-relative descriptors and Mahalanobis distance. In a three-link arm task, orbit success rose from 0.49 under uniform sampling to 0.69, while a labeled oracle reached 0.74; this was one fixed log, and the comparison changes both windows and representation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.