Policy Gradient with Expected Quadratic Utility Maximization: A New Mean-Variance Approach in Reinforcement Learning.
Masahiro KatoKei NakagawaPublished in: CoRR (2020)
Keyphrases
- policy gradient
- utility maximization
- utility function
- reinforcement learning
- actor critic
- reinforcement learning algorithms
- function approximation
- optimal control
- gradient method
- pairwise
- decision makers
- approximation methods
- partially observable markov decision processes
- temporal difference
- state action
- function approximators
- average reward
- reinforcement learning methods
- learning algorithm
- model free
- decision problems
- probability distribution
- variance reduction
- objective function
- computational complexity
- markov decision processes
- dynamic programming
- multi agent
- state space
- optimal policy
- partially observable
- long run