Performance-Weighed Policy Sampling for Meta-Reinforcement Learning.
Ibrahim AhmedMarcos Quiñones-GrueiroGautam BiswasPublished in: CoRR (2020)
Keyphrases
- reinforcement learning
- optimal policy
- approximate policy iteration
- policy search
- action selection
- markov decision process
- markov decision processes
- policy iteration
- markov decision problems
- actor critic
- partially observable environments
- state and action spaces
- action space
- function approximation
- reward function
- policy gradient
- reinforcement learning algorithms
- temporal difference
- control policy
- control policies
- monte carlo
- reinforcement learning problems
- continuous state
- state action
- policy evaluation
- approximate dynamic programming
- partially observable
- random sampling
- function approximators
- temporal difference learning
- rl algorithms
- average reward
- partially observable markov decision processes
- asymptotically optimal
- policy making
- model free
- infinite horizon
- partially observable domains
- learning algorithm
- machine learning
- control problems
- state space
- meta level
- sample size
- decision problems
- sampling strategy
- dynamic programming
- agent learns
- inverse reinforcement learning
- transition model
- long run
- sampling methods
- reinforcement learning methods
- continuous state spaces
- finite state
- sampling algorithm
- average cost
- robotic control
- learning process
- multi agent
- model free reinforcement learning