RAAM: The Benefits of Robustness in Approximating Aggregated MDPs in Reinforcement Learning.
Marek PetrikDharmashankar SubramanianPublished in: NIPS (2014)
Keyphrases
- reinforcement learning
- markov decision processes
- state space
- optimal policy
- function approximation
- markov decision process
- reinforcement learning algorithms
- model free
- policy iteration
- continuous state and action spaces
- partially observable
- dynamic programming
- action space
- policy search
- state and action spaces
- model based reinforcement learning
- reward function
- learning algorithm
- multi agent
- control problems
- learning process
- policy evaluation
- machine learning
- action sets
- optimal control
- finite state
- temporal difference
- finite horizon
- markov chain
- control policy
- reinforcement learning methods
- continuous state
- sufficient conditions
- linear programming
- stochastic domains
- transfer learning