Fast reinforcement learning with generalized policy updates.
André BarretoShaobo HouDiana BorsaDavid SilverDoina PrecupPublished in: Proc. Natl. Acad. Sci. USA (2020)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- action selection
- markov decision process
- markov decision processes
- action space
- function approximation
- actor critic
- reinforcement learning problems
- state space
- policy gradient
- control policy
- partially observable environments
- reinforcement learning algorithms
- control policies
- partially observable
- reward function
- function approximators
- markov decision problems
- state and action spaces
- policy iteration
- approximate dynamic programming
- policy evaluation
- partially observable markov decision processes
- learning process
- temporal difference
- transition model
- state action
- rl algorithms
- long run
- continuous state
- optimal control
- partially observable domains
- decision problems
- model free
- database
- finite state
- multi agent reinforcement learning
- average reward
- continuous state spaces
- dynamical systems
- transfer learning
- asymptotically optimal
- robotic control
- multi agent