From free energy to expected energy: Improving energy-based value function approximation in reinforcement learning.
Stefan ElfwingEiji UchibeKenji DoyaPublished in: Neural Networks (2016)
Keyphrases
- markov random field
- free energy
- belief propagation
- energy minimization
- reinforcement learning
- graph cuts
- temporal difference
- state space
- approximate inference
- pairwise
- fixed point
- function approximation
- state action
- reinforcement learning algorithms
- markov decision processes
- dynamic programming
- neural network
- maximum likelihood
- upper bound
- learning algorithm