AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Information Optimization and Transferable State Abstractions in Deep Reinforcement Learning

Diego Gomez, Nicanor Quijano, Luis Felipe Giraldo

TPAMI 2022 · Reinforcement learning › Visual and representation RL · approach: Learned state and action representations

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 information concepts with TRMC instead of SACSAC · AntSquareWallepisodes to 50% of max return: 93 -> 16Improved
use information concepts with TRMC instead of SACSAC · AntAvoidepisodes to 80% of max return: 180 -> 34Improved
use information concepts with TRMC instead of SACSAC · AntGatherepisodes to 80% of max return: 363 -> 541Hurt
use information concepts instead of SAC policySAC · AntSquareWallasymptotic performance gap: 69.0Hurt
use mutual-information metric instead of maximum likelihood metric (ablation)TRMC - LikelihoodConcepts · AntGatherJ_ap: 2095.2 -> 965.3Improved
use mutual-information metric instead of maximum likelihood metric (ablation)TRMC - LikelihoodConcepts · AntAvoidepisodes to 50% of max return: Not reached -> 19Improved
condition the information metric on task context (ablation)TRMC - InfoConcepts (no context) · AntGatherJ_ap: 3250.5 -> 965.3Improved
condition the information metric on task context (ablation)TRMC - InfoConcepts (no context) · AntSquareTrackepisodes to 80% of max return: Not reached -> 32Improved
condition the information metric on task context (ablation)TRMC - InfoConcepts (no context) · AntSquareWallepisodes to 50% of max return: 25 -> 16Improved
transfer all Reptile weights instead of only fixed convolutional features (ablation)SAC - Reptile (visual features) · AntSquareWallJ_ap: 151.4 -> 48.0Improved
use learned concepts as behavior prior on unseen mazeSAC · AntCrossMazeconvergence speed: 2 times faster to reach 85% of max returnImproved
transfer concepts to unseen LunarLander gravity tasksDDQN · LunarLandersample efficiencyMixed

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