RL-squared and recurrent deep meta-reinforcement learning

What changed

Learns history-dependent recurrent policies across task distributions so fixed network weights support fast within-task adaptation through changing recurrent state.

Scientific scope

Learns history-dependent recurrent policies across task distributions so fixed network weights support fast within-task adaptation through changing recurrent state. The contemporaneous recurrent meta-reinforcement-learning approaches have separate source bylines and GRU/TRPO versus LSTM/actor-critic implementations. Their trial and episode-reset conventions matter. Near-optimal performance on the tested bandits and navigation tasks does not establish optimality on arbitrary new tasks.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. The contemporaneous recurrent meta-reinforcement-learning approaches have separate source bylines and GRU/TRPO versus LSTM/actor-critic implementations. Their trial and episode-reset conventions matter. Near-optimal performance on the tested bandits and navigation tasks does not establish optimality on arbitrary new tasks.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• RL$^2$: Fast Reinforcement Learning via Slow Reinforcement Learning
• Learning to reinforcement learn

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction