Reward-predictive representations generalize across tasks in reinforcement learning.
Lucas LehnertMichael L. LittmanMichael J. FrankPublished in: PLoS Comput. Biol. (2020)
Keyphrases
- reinforcement learning
- transfer learning
- reinforcement learning algorithms
- function approximation
- multi agent environments
- markov decision processes
- state space
- complex domains
- partially observable environments
- eligibility traces
- reinforcement learning agents
- function approximators
- model free
- supervised learning
- machine learning
- optimal policy
- robotic control
- multi task
- dynamic programming
- control policy
- state action
- multiple representations
- adaptation process
- policy gradient
- higher level
- markov chain