AI Paper Atlas

Research map / Reinforcement learning / Imitation learning

When Demonstrations meet Generative World Models: A Maximum Likelihood Framework for Offline Inverse Reinforcement Learning

Siliang Zeng, Chenliang Li, Alfredo Garcia, Mingyi Hong

NeurIPS 2023 · Reinforcement learning › Imitation learning · approach: Reward learning from demonstrations · 1 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
train world model on medium-expert instead of medium transitionsOffline ML-IRL · D4RL halfcheetah, 5,000 demosaverage episode reward: 7640.73 -> 11231.40Improved
train world model on medium-replay instead of medium transitionsOffline ML-IRL · D4RL halfcheetah, 5,000 demosaverage episode reward: 7640.73 -> 9236.84Improved
Offline ML-IRL versus CLARE (also model-based offline IRL)Offline ML-IRL · D4RL halfcheetah medium-exp, 5,000 demosaverage episode reward: 622.79 -> 11231.40Improved
Offline ML-IRL versus BC on halfcheetah mediumOffline ML-IRL · D4RL halfcheetah medium, 5,000 demosaverage episode reward: 4471.72 -> 7640.73Improved
Offline ML-IRL versus ValueDICE on hopper mediumOffline ML-IRL · D4RL hopper medium, 5,000 demosaverage episode reward: 3073.16 -> 2453.25Hurt
reduce expert demonstrations from 5,000 to 1,000Offline ML-IRL · D4RL halfcheetah mediumaverage episode reward: 7640.73 -> 7690.30No clear effect
reduce expert demonstrations from 5,000 to 1,000BC · D4RL halfcheetah mediumaverage episode reward: 4471.72 -> 2799.40Hurt
reduce expert demonstrations from 5,000 to 1,000ValueDICE · D4RL walker2daverage episode reward: 3191.47 -> 1794.97Hurt
increase expert demonstrations from 5,000 to 10,000Offline ML-IRL · D4RL walker2d medium-expaverage episode reward: 4201.40 -> 4363.54No clear effect
increase expert demonstrations from 5,000 to 10,000BC · D4RL hopperaverage episode reward: 2801.19 -> 3236.70Improved
transfer reward learned on medium-expert to medium-replay and run MOPOOffline ML-IRL · D4RL medium-replay offline RLaverage episode rewardNo clear effect
exclude IQ-Learn from comparisonIQ-Learn · D4RLaverage episode rewardHurt

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