AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

Towards mental time travel: a hierarchical memory for reinforcement learning agents

Andrew Kyle Lampinen, Stephanie C.Y. Chan, Andrea Banino, Felix Hill

NeurIPS 2021 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Language-grounded reinforcement agents · 5 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 HCAM memory instead of Gated TransformerXLRL agent · Ballet 2 dances, delay 16% correct: 96.9 -> 99.8Improved
use HCAM memory instead of Gated TransformerXLRL agent · Ballet 8 dances, delay 16% correct: 65.7 -> 98.1Improved
use HCAM memory instead of Gated TransformerXLRL agent · Ballet 8 dances, delay 48% correct: 49.2 -> 97.2Improved
use HCAM memory instead of LSTMRL agent · Ballet 8 dances, delay 16% correct: 25.2 -> 98.1Improved
use HCAM memory instead of TrXLRL agent · Object permanence, no delay% correct: 82.2 -> 96.7Improved
use HCAM memory instead of TrXLRL agent · Object permanence, long delay% correct: 46.1 -> 91.7Improved
train only on long object-permanence tasks (no short scaffold) with HCAMHCAM · Object permanence, long delay% correct: 91.7 -> 82.9Mixed
use HCAM instead of TrXL for extrapolation to 10 distractorsRL agent · Rapid word learning% correct: 76.7 -> 93.0Improved
use HCAM instead of TrXL for recall across 4 episodesRL agent · Rapid word learning% correct: 49.3 -> 82.8Improved
use HCAM instead of TrXL for 2 episodes with 5 distractors eachRL agent · Rapid word learning% correct: 48.7 -> 71.1Improved
use HCAM instead of TrXLRL agent · One-Shot StreetLearnaverage reward: 19.9 -> 26.8Improved
select a single memory chunk instead of top-k (ablation)HCAM · Ballet% correctNo clear effect

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