AI Paper Atlas

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

Improving and Accelerating Offline RL in Large Discrete Action Spaces with Structured Policy Initialization

Matthew Landers, Taylor W. Killian, Thomas Hartvigsen, Afsaneh Doryab

ICLR 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Model-based RL with guarantees · 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
decouple structure learning from control via pre-trained frozen ASMSAINT-style policy with IQL · DM Control overallaverage return: 572.1 -> 594.1Improved
add ASM pre-training before policy learningSAINT-style policy with IQL · DM Control overalltime to target (min): 845.0 -> 223.3Improved
add structured pre-training on medium (near-homogeneous) dataSAINT-style policy with IQL · DM Control mediumaverage return: 343.0 -> 345.2No clear effect
increase action cardinality from 3 to 30 bins (ablation)SPIN vs SAINT · dog-trot medium-expertaverage return: SPIN advantage over SAINT 647.0 vs 635.1 -> 703.9 vs 562.5Improved
train ASM longer (0 to 20+ epochs) (ablation)SPIN · DM Control medium-expertfinal returnImproved
pre-train ASM instead of random init (ablation)ASM linear probe · dog-trot probeexact-match accuracy: 0.10% -> 4.52%Improved
replace Transformer policy with distilled MLP (ablation)SPIN · DM Control medium-expertaverage return: 753.2 -> 748.8No clear effect
use variational or contrastive pre-training instead of masked modeling (ablation)SPIN · DM Control medium-expertreturn: cheetah 651.1 -> 607.5 (VAM), 607.3 (CAM)Hurt
use SPIN with AWAC objectiveSAINT with AWAC · DM Control medium-expertaverage return: 716.5 -> 764.2Improved
use SPIN with BCQ objectiveSAINT with BCQ · DM Control medium-expertaverage return: 760.8 -> 773.1Improved
use SPIN on MazeF-IQL · Mazeaverage return: 99.0 -> 99.4No clear effect
action-centric pre-training instead of trajectory-centric RePreM (ablation)SPIN · DM Control random-medium-expertaverage return: 372.2 -> 499.2Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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