SQOG adds a noise-based smoothing loss that pulls out-of-distribution action values toward neighboring in-sample values, improving Q-estimation and offline RL performance.
Flows for simultaneous manifold learning and density estimation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Offline RL with Smooth OOD Generalization in Convex Hull and its Neighborhood
SQOG adds a noise-based smoothing loss that pulls out-of-distribution action values toward neighboring in-sample values, improving Q-estimation and offline RL performance.