AI Paper Atlas

Research map / Reinforcement learning / MDPs and planning

Corruption-Robust Offline Reinforcement Learning with General Function Approximation

Chenlu Ye, Rui Yang, Quanquan Gu, Tong Zhang

NeurIPS 2023 · Reinforcement learning › MDPs and planning · approach: Function approximation in RL theory · 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
add uncertainty weighting to MSG (ablation)MSG · 12 corrupted D4RL-style tasks (average)average return: 1083.6 -> 2212.8Improved
add uncertainty weighting to MSG (ablation)MSG · Halfcheetah adversarial dynamicsaverage return: -87.3 -> 4144.3Improved
add uncertainty weighting to MSG (ablation)MSG · Halfcheetah random rewardaverage return: 4339.5 -> 7299.9Improved
add uncertainty weighting to MSG (ablation)MSG · Halfcheetah random dynamicsaverage return: 212.9 -> 1425.0Improved
add uncertainty weighting to MSG (ablation)MSG · Walker2d random rewardaverage return: 539.1 -> 2189.7Improved
add uncertainty weighting to MSG (ablation)MSG · Walker2d random dynamicsaverage return: 2122.8 -> 2278.9Improved
add uncertainty weighting to MSG (ablation)MSG · Walker2d adversarial rewardaverage return: 605.8 -> 1433.4Improved
add uncertainty weighting to MSG (ablation)MSG · Hopper random rewardaverage return: 1599.6 -> 2021.1Improved
add uncertainty weighting to MSG (ablation)MSG · Hopper adversarial dynamicsaverage return: 717.7 -> 931.2Improved
use UWMSG instead of SAC-N under random reward corruptionUWMSG · Halfcheetah random rewardaverage return: 7357.5 -> 7299.9No clear effect
use UWMSG instead of EDAC under adversarial dynamicsUWMSG · Halfcheetah adversarial dynamicsaverage return: 374.0 -> 4144.3Improved
use UWMSG instead of SAC-N on Hopper adversarial rewardUWMSG · Hopper adversarial rewardaverage return: 111.8 -> 751.4Improved

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