AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Active Policy Optimization for Individualized Dosing via Gradient Variance Minimization

Yi Wan, Xin Wang, Huanhuan Chen

ICML 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Trajectory reuse in policy gradients

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 gradient variance instead of function value as selection criterion (ablation)GVALID · HNL (N=300)policy suboptimality R: 0.26 -> 0.15Improved
use gradient variance instead of function value as selection criterion (ablation)GVALID · CSC (N=300)policy suboptimality R: 0.77 -> 0.48Improved
use gradient variance instead of function value as selection criterion (ablation)GVALID · News (N=300)policy suboptimality R: 0.30 -> 0.07Improved
use gradient variance instead of function value as selection criterion (ablation)GVALID · News (N=100)policy suboptimality R: 0.55 -> 0.16Improved
jointly optimize covariates x and dose t rather than random x (ablation)GVALID · HNL (N=300)policy suboptimality R: 0.25 -> 0.15Improved
jointly optimize covariates x and dose t rather than random x (ablation)GVALID · CSC (N=300)policy suboptimality R: 1.09 -> 0.48Improved
jointly optimize covariates x and dose t rather than random x (ablation)GVALID · SWV (N=300)policy suboptimality R: 0.35 -> 0.33No clear effect
jointly optimize covariates x and dose t rather than random x (ablation)GVALID · News (N=100)policy suboptimality R: 0.18 -> 0.16Improved
use moderate batch size instead of 1 or 20 (ablation)GVALID · HNL and SWVpolicy suboptimality RImproved
use GVALID instead of random samplingRAND baseline · STS (N=350, x1e-2)policy suboptimality R: 0.78 -> 0.17Improved
use GVALID instead of MaxVarMaxVar baseline · HNL (N=350)policy suboptimality R: 0.18 -> 0.12Improved
use GVALID instead of ABC3ABC3 baseline · STS (N=350, x1e-2)policy suboptimality R: 0.17 -> 0.17No 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