PoPS: Policy Pruning and Shrinking for Deep Reinforcement Learning.
Dor LivneKobi CohenPublished in: CoRR (2020)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- markov decision process
- reinforcement learning problems
- action selection
- state and action spaces
- partially observable
- reinforcement learning algorithms
- actor critic
- function approximators
- state space
- action space
- partially observable environments
- control policy
- function approximation
- policy gradient
- reward function
- rl algorithms
- control policies
- policy iteration
- markov decision problems
- model free
- markov decision processes
- policy evaluation
- dynamic programming
- search space
- partially observable domains
- continuous state spaces
- approximate dynamic programming
- finite state
- partially observable markov decision processes
- infinite horizon
- state dependent
- state action
- pruning method
- transition model
- machine learning
- neural network
- optimal control
- inverse reinforcement learning
- policy gradient methods
- learning algorithm
- multi agent
- agent learns
- reinforcement learning methods
- temporal difference
- asymptotically optimal
- deep learning
- average reward
- temporal difference learning
- learning problems
- transfer learning
- pruning algorithms
- allocation policy
- agent receives
- continuous state