AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Self-Paced Context Evaluation for Contextual Reinforcement Learning

Theresa Eimer, André Biedenkapp, Frank Hutter, Marius Lindauer

ICML 2021 · Reinforcement learning › Visual and representation RL · approach: In-context and adaptive RL · 3 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
Use value-function progress to choose curriculum instances instead of round-robin orderingPPO on Ant · Anttraining steps to reach reward around 11: 10× as many stepsImproved
Use value-function progress to choose curriculum instances instead of round-robin orderingDQN · CartPolemean performance per episode: 25 -> around 150Improved
Use value-function progress to choose curriculum instances instead of round-robin orderingSPaCE on BallCatching · BallCatchingtraining speed: a factor of at least 10Improved
Use value-function progress to choose curriculum instances instead of round-robin orderingSPaCE on PointMass · PointMass, uniformly sampled instancesfinal performanceImproved
Use value-function progress to choose curriculum instances instead of round-robin orderingSPaCE on mazes · 5x5 mazesepisodes to generalize across mazes: 10 episodes vs several hundred episodesImproved
Target the entire instance space rather than focusing on generated hard instancesSPaCE vs SPDRL · PointMass, uniformly sampled test instanceslearning speed and final performanceImproved
Replace value-progress selection with context-space distance (ablation)SPaCE (context SPaCE) · PointMasslearning speed and overall performanceNo clear effect
Use a larger curriculum increment κ across the tested values (ablation)SPaCE on PointMass · PointMass test, after $10^6$ stepsmean rewardNo clear effect
Vary the performance threshold η across the tested values (ablation)SPaCE on PointMass · PointMass test, after $10^6$ stepsmean rewardNo clear effect
Multiply the training-set size instead of adding instances incrementally (ablation)SPaCE · PointMasstraining speedHurt
Train on only 25 or 50 instances rather than larger sets (ablation)SPaCE · PointMass testtest performanceHurt
Use SPDRL's hard-instance distribution rather than a space-covering instance set (ablation)SPDRL · PointMass, hard instancesperformance over training: first ~200,000 stepsMixed

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