AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Unsupervised Behavior Extraction via Random Intent Priors

Hao Hu, Yiqin Yang, Jianing Ye, Ziqing Mai, Chongjie Zhang

NeurIPS 2023 · Reinforcement learning › Offline RL · approach: Diffusion models for offline decision making · 0 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
Use random intent priors instead of behavior cloning to extract offline behaviorsUBER policy reuse · Antmaze medium-play, 100k stepsnormalized score: 0.0±0.0 -> 91.9±2.9Improved
Use random intent priors instead of behavior cloning to extract offline behaviorsUBER policy reuse · Antmaze large-play, 100k stepsnormalized score: 0.0±0.0 -> 62.8±2.6Improved
Use random intent priors instead of average-reward offline behavior extraction (ablation)UBER · Mujoco downstream tasksonline performanceImproved
Use policy expansion (PEX) rather than critic-guided policy reuse (CUP) (ablation)UBER · Downstream tasksonline performanceImproved
Use random intent priors rather than behavior cloningBehavior extraction from reward-free offline data · Return-distribution entropy across D4RL tasksreturn-distribution entropy: DATASET: 1.21, 1.47, 0.86, 0.71, 1.97, 0.82; UBER: 2.01, 1.92, 2.69, 2.45, 2.32, 1.11Improved
Use random intent priors rather than behavior cloningBehavior extraction from reward-free offline data · Antmaze medium, largeonline performanceImproved

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