AI Paper Atlas

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

Direct Advantage Estimation

Hsiao-Ru Pan, Nico Gürtler, Alexander Neitz, Bernhard Schölkopf

NeurIPS 2022 · Reinforcement learning › Policy gradient and actor-critic · approach: Temporal-difference and discounting theory · 2 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
replace GAE with DAEPPO · MinAtar (5 envs, baseline network)Overall/Last score: 6.6 -> 10.5Improved
replace GAE with DAEPPO · ALE (baseline network)environments won: 10 -> 32Improved
replace GAE with DAE, Wide networkPPO · ALEenvironments won (Overall): 8 -> 35Improved
replace GAE with DAE, Deep networkPPO · ALEenvironments won (Overall): 6 -> 35Improved
widen network layersDAE · MinAtar BreakoutLast score: 844.7 -> 2409.5Improved
widen network layersGAE · MinAtar BreakoutLast score: 168.2 -> 637.7Improved
widen network layersDAE · MinAtar SpaceInvadersLast score: 1675.3 -> 11743.3Improved
widen network layersGAE · MinAtar SpaceInvadersLast score: 504.2 -> 1058.0Improved
replace GAE with DAEPPO · synthetic environmentMSE of advantage estimateImproved
replace GAE with DAEPPO · synthetic environment, early trainingexpected returnHurt
replace GAE with DAEPPO · synthetic environment, late trainingexpected returnImproved
replace DAE loss with Indirect (separate Q and V) (ablation)DAE · MinAtarnormalized scoreHurt

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