Policy Feedback in Deep Reinforcement Learning to Exploit Expert Knowledge.
Federico EspositiAndrea BonariniPublished in: LOD (1) (2020)
Keyphrases
- expert knowledge
- reinforcement learning
- optimal policy
- policy search
- expert systems
- markov decision process
- knowledge engineering
- domain knowledge
- fuzzy logic
- action selection
- partially observable environments
- action space
- function approximation
- domain experts
- markov decision processes
- policy gradient
- reinforcement learning problems
- learned knowledge
- prior knowledge
- approximate dynamic programming
- control policy
- reinforcement learning algorithms
- bayesian networks
- markov decision problems
- knowledge base
- policy iteration
- control policies
- function approximators
- actor critic
- partially observable
- policy evaluation
- state and action spaces
- reward function
- state action
- rl algorithms
- learning algorithm
- state space
- partially observable markov decision processes
- state dependent
- reward signal
- continuous state spaces
- finite state
- temporal difference
- model free
- knowledge elicitation
- relevance feedback
- transition model
- domain specific
- inverse reinforcement learning
- continuous state
- neural network
- partially observable domains
- reinforcement learning methods