Reproduce “Downstream-Aware Context Selection for Online 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
Ruihan A. Li, Shangtong Zhang, Rohan Chandra
The method trims histories in in-context reinforcement learning by predicting the downstream effect of removing interaction records, then selecting a history size that meets a threshold. On held-out in-distribution SUMO driving with GPT-4o, it reduced average total tokens from 218K to 162K, or 25.7%, while measured Safety, Comfort, and Efficiency scores remained comparable; this finding is limited to that setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.