AI Paper Atlas

Research map / Reinforcement learning / Imitation learning

Imitating Past Successes can be Very Suboptimal

Benjamin Eysenbach, Soumith Udatha, Ruslan Salakhutdinov, Sergey Levine

NeurIPS 2022 · Reinforcement learning › Imitation learning · approach: Behavior cloning theory and models · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add normalization (remove averaging step) to OCBCOCBC/GCSL · GCSL benchmark original tasksnormalized return: +3\%No clear effect
add normalization to OCBC under skewed action distributionOCBC/GCSL · GCSL benchmark skewed actionsnormalized return: +149\%Improved
add normalization to OCBC under biased goal distributionOCBC/GCSL · GCSL benchmark biased goalsnormalized return: +62\%Improved
add normalization to iterated OCBCiterated OCBC · 3-action banditpolicy optimality for tasks e2 and e3Improved
add normalization to OCBC (skewed commanded goals 9:1)OCBC · 2D navigation, orange goalsteps to reach goalImproved
add normalization to OCBCOCBC · 2D navigation, blue goalsteps to reach goalNo clear effect
label many states as failures (discard trials) in OCBC (ablation)OCBC · 1D gridworld, high-data regimeaverage rewardImproved
label many states as failures (discard trials) in OCBC (ablation)OCBC · 1D gridworld, low-data regimeaverage rewardHurt
filtered relabeling with importance-weight clipping epsilon (ablation)Normalized OCBC · Pointmass rooms (random and biased-goal datasets)performanceNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy