Ensuring Monotonic Policy Improvement in Entropy-regularized Value-based Reinforcement Learning.
Lingwei ZhuTakamitsu MatsubaraPublished in: CoRR (2020)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- action selection
- markov decision process
- policy evaluation
- least squares
- actor critic
- function approximation
- partially observable environments
- policy iteration
- state space
- markov decision processes
- control policy
- approximate dynamic programming
- reinforcement learning algorithms
- action space
- control policies
- partially observable
- dynamic programming
- optimal control
- temporal difference
- machine learning
- state and action spaces
- markov decision problems
- information theoretic
- state action
- function approximators
- decision problems
- information theory
- model free
- policy gradient
- mutual information
- reinforcement learning problems
- reward function
- average reward
- rl algorithms
- partially observable markov decision processes
- continuous state
- supervised learning
- information entropy
- multi agent
- neural network