Reward-Free Policy Space Compression for Reinforcement Learning.
Mirco MuttiStefano Del ColMarcello RestelliPublished in: AISTATS (2022)
Keyphrases
- reinforcement learning
- action space
- optimal policy
- partially observable environments
- reward function
- action selection
- policy gradient
- policy search
- reinforcement learning algorithms
- average reward
- markov decision processes
- state action
- markov decision process
- state space
- total reward
- function approximators
- control policy
- partially observable
- function approximation
- eligibility traces
- actor critic
- control policies
- policy evaluation
- compression algorithm
- reinforcement learning problems
- space time
- approximate dynamic programming
- state dependent
- inverse reinforcement learning
- markov decision problems
- agent receives
- learning algorithm
- image compression
- temporal difference
- decision problems
- reinforcement learning methods
- multi agent
- search space
- agent learns
- data compression
- transition model
- model free
- policy iteration
- continuous state
- rl algorithms
- long run
- expected reward
- machine learning