Genetic Soft Updates for Policy Evolution in Deep Reinforcement Learning.
Enrico MarchesiniDavide CorsiAlessandro FarinelliPublished in: ICLR (2021)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- action selection
- markov decision process
- grammatical evolution
- function approximators
- markov decision processes
- function approximation
- genetic algorithm
- reinforcement learning problems
- control policy
- control policies
- partially observable environments
- reinforcement learning algorithms
- partially observable
- approximate dynamic programming
- actor critic
- policy gradient
- state and action spaces
- finite state
- reward function
- state space
- continuous state
- state action
- model free
- policy iteration
- machine learning
- long run
- decision problems
- markov decision problems
- continuous state spaces
- action space
- policy evaluation
- agent learns
- average reward
- state dependent
- temporal difference
- infinite horizon
- genetic programming
- dynamic programming
- learning process
- learning algorithm
- update operations
- neural network
- transition model
- supervised learning
- population genetics
- model free reinforcement learning
- exploration exploitation tradeoff