AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Continual Auxiliary Task Learning

Matthew K McLeod, Chunlok Lo, Matthew Kyle Schlegel, Andrew Jacobsen, Raksha Kumaraswamy, Martha White, Adam M White

NeurIPS 2021 · Reinforcement learning › Visual and representation RL · approach: Learned state and action representations · 2 citations (OpenAlex)

Read the paper

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 SF-NR instead of Tree Backup for GVF learningContinual Auxiliary Task Learning · Tabular TMazeprediction RMSEImproved
Use SF-NR instead of Tree Backup for GVF learningContinual Auxiliary Task Learning · Continuous TMazeprediction RMSEImproved
Use GPI instead of Sarsa for behavior controlContinual Auxiliary Task Learning · Tabular TMazeprediction RMSENo clear effect
Use GPI instead of Sarsa for behavior controlContinual Auxiliary Task Learning · Continuous TMazeprediction RMSEImproved
Use GPI instead of a random behavior policyContinual Auxiliary Task Learning · Mountain CarGVF prediction RMSEImproved
Use replay for the stationary successor-feature component (ablation)SF-NR · Continuous TMazeprediction RMSEImproved
Add replay to Tree Backup learning (ablation)Tree Backup · Continuous TMazeprediction RMSEHurt
Add interest functions to focus predictions on relevant states (ablation)GVF learners · Open 2D Worldprediction RMSEImproved
Use emphatic reweighting with interest instead of interest alone (ablation)GVF learners · Open 2D Worldprediction RMSEMixed
Use prior corrections to reweight learning toward the target policy (ablation)GVF learners · Off-policy learning with exploratory behaviorlearning effectivenessHurt
Use reward features that generalize excessively across states (ablation)SF-NR · Continuous TMazeprediction RMSEHurt
Learn values directly with LSTD instead of separating successor features and cumulant learningGVF learners · Tabular TMazeprediction RMSEHurt

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