Reproduce “Asking Easy Questions: A User-Friendly Approach to Active Reward 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
Erdem Bıyık, Malayandi Palan, Nicholas C. Landolfi, Dylan P. Losey, Dorsa Sadigh
The information-gain method selects preference questions balancing uncertainty about user answers across reward hypotheses with uncertainty in the user’s response under a given reward. In simulations it improved reward-weight alignment over volume removal across four environments with strict and weak queries. Participants rated its questions easier in all tested settings (p < 0.005), though ease depends on the human-choice model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.