AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Offline RL with No OOD Actions: In-Sample Learning via Implicit Value Regularization

Haoran Xu, Li Jiang, Jianxiong Li, Zhuoran Yang, Zhaoran Wang, Victor Wai Kin Chan, Xianyuan Zhan

ICLR 2023 · Reinforcement learning › Offline RL · approach: Behavior regularization for offline RL · 4 citations (OpenAlex)

Read the paper

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 SQL instead of CQL for in-sample value learningOffline RL · AntMaze medium and large small-data tasksnormalized return: 65.2 -> 75.1; 9.3 -> 24.2; 6.3 -> 36.7; 0 -> 34.2Improved
Use SQL instead of CQL for in-sample value learningOffline RL · AntMaze medium and large small-data tasksBellman error: 13.1 -> 1.6; 64.4 -> 1.9; 30.6 -> 1.3; 300.5 -> 2.6Improved
Add SQL's sparsity term to value learningSQL versus IQL · Noisy mixed expert/random MuJoCo datasetsnormalized scoreImproved
Vary SQL regularization parameter alpha (ablation)SQL · MuJoCo meannormalized score: 30.4 (alpha 0.5) -> 79.0 (alpha 10)Mixed
Vary SQL regularization parameter alpha (ablation)SQL · AntMaze meannormalized score: 18.9 (alpha 0.2) -> 64.5 (alpha 0.5) -> 7.5 (alpha 2.0)Mixed
Increase SQL alpha from 0.5 to 1 (ablation)SQL · halfcheetah-mediumnormalized score: 48.3 -> 48.7Improved
Increase SQL alpha from 0.5 to 2 (ablation)SQL · walker2d-mediumnormalized score: 22.3 -> 84.2Improved
Increase SQL alpha from 0.5 to 2 (ablation)SQL · AntMaze medium-playnormalized score: 0.0 -> 80.2Improved
Use SQL instead of IQLOffline RL · D4RL AntMaze and Kitchen benchmarksnormalized scoreImproved
Use SQL instead of IQLOffline RL · D4RL MuJoCo tasksnormalized scoreNo clear effect
Use SQL instead of IQL in Four RoomsOffline RL with random behavior data · Four Roomspolicy successImproved
Use SQL instead of CQL in small-data AntMazeOffline RL · AntMaze small-data tasksnormalized returnImproved

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