AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Budgeting Counterfactual for Offline RL

Yao Liu, Pratik Chaudhari, Rasool Fakoor

NeurIPS 2023 · Reinforcement learning › Offline RL · approach: Simple practical offline RL · 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
Use BCOL (SAC) instead of CQLSAC-based offline RL · D4RL totalnormalized score: 1065.9 -> 1142.0Improved
Use BCOL (SAC) instead of CDCSAC-based offline RL · D4RL totalnormalized score: 1032.5 -> 1142.0Improved
Use BCOL (TD3) instead of TD3+BCTD3-based offline RL · D4RL totalnormalized score: 840.2 -> 873.3Improved
Use BCOL (SAC) instead of IQLoffline RL · D4RL AntMaze totalnormalized score: 378.0 -> 396.0Improved
Use BCOL (SAC) instead of ARMORmodel-based offline RL · D4RL MuJoCo totalnormalized score: 788.6 -> 746.0Hurt
Use BCOL (SAC) instead of CDC on MuJoCoSAC-based offline RL · D4RL MuJoCo totalnormalized score: 699.0 -> 746.0Improved
Use BCOL (TD3) instead of TD3+BC on MuJoCoTD3-based offline RL · D4RL MuJoCo totalnormalized score: 669.2 -> 685.6Improved
Remove the budget constraint (B to infinity) (ablation)BCOL · D4RL MuJoCo and AntMazepercent difference in scoreHurt
Assign B decisions randomly to the policy (no planning in training or testing) (ablation)BCOL · D4RLpercent difference in scoreHurt
Select B policy steps randomly at test time (no test-time planning) (ablation)BCOL · D4RLpercent difference in scoreHurt
Budget of only B=1 counterfactual stepBCOL (SAC) · D4RL MuJoConormalized scoreNo clear effect
Budget of 10 steps out of 1000BCOL (SAC) · D4RL MuJoConormalized scoreImproved

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