Policy Learning research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

HIV dataset

Harm-unaware policy

6.063Estimated Expected Outcome

May 26, 2026

ELAIA AKI-alert trial 1.0

Never alert

0.905Certified Value

Apr 3, 2026

Case 2 Simulated Dataset

Fair CATE

12.9UF Score

Mar 6, 2026

Case 1 simulation

AF-IDR

0.027UF

Mar 6, 2026

CASdatasets Counterfactual (test)

Optimal

-9.38V(pi)

Feb 26, 2026

CASdatasets Equal Opportunity (test)

Optimal

-9.38V(pi)

Feb 26, 2026

Meta-World 15 tasks

ACT-JEPA

92Average Task Success Rate

Mar 19, 2026

ManiSkill 5 tasks

ACT-JEPA

36Average Task Success Rate

Mar 19, 2026

Push-T 1 task

ACT-JEPA

41Average Task Success Rate

Mar 19, 2026

Tabular Discounted MDPs Generative Setting

MB-RS-QVI

2Upper Bound (Complexity)

May 20, 2026

MNIST action-exclusive feedback (test)

—

—Primary metric

Feb 26, 2026

MNIST action-inclusive feedback (test)

—

—Primary metric

Feb 26, 2026