AI Paper Atlas

Research map / Reinforcement learning / MDPs and planning

Seeing is not Believing: Robust Reinforcement Learning against Spurious Correlation

Wenhao Ding, Laixi Shi, Yuejie Chi, Ding Zhao

NeurIPS 2023 · Reinforcement learning › MDPs and planning · approach: POMDPs and confounded RL · 2 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 causal graph G_phi (vs removing it) (ablation)RSC-SAC · Behavior shiftednormalized reward: 0.51 -> 1.02Improved
add causal graph G_phi (vs removing it) (ablation)RSC-SAC · Lift shiftednormalized reward: 0.79 -> 0.98Improved
add transition model P_theta (vs removing it) (ablation)RSC-SAC · Behavior shiftednormalized reward: 0.41 -> 1.02Improved
add transition model P_theta (vs removing it) (ablation)RSC-SAC · Crossing shiftednormalized reward: 0.89 -> 1.04Improved
add confounder distribution P^c (vs removing it) (ablation)RSC-SAC · Behavior shiftednormalized reward: 0.66 -> 1.02Improved
add confounder distribution P^c (vs removing it) (ablation)RSC-SAC · Lift shiftednormalized reward: 0.90 -> 0.98Improved
add L1 sparsity regularization to SACSAC · distraction correlation tasksnormalized rewardImproved
add L1 sparsity regularization to SACSAC · composition correlation tasksnormalized rewardNo clear effect
use traditional robust RL (RMDP-G, RMDP-U, ATLA) outside their uncertainty setSAC · shifted environmentsnormalized rewardHurt
RSC-SAC vs RMDP-0.01 under Gaussian noise 0.1RSC-SAC · Lift-0.01normalized reward: 0.96 vs 0.94No clear effect
RSC-SAC vs RMDP-0.1 in nominal LiftRSC-SAC · Lift-0normalized reward: 0.85 -> 0.96Improved
increase augmentation ratio beta to 1% (ablation)RSC-SAC · nominal and shiftednormalized rewardNo clear effect

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