The Mirage of Action-Dependent Baselines in Reinforcement Learning.
George TuckerSurya BhupatirajuShixiang GuRichard E. TurnerZoubin GhahramaniSergey LevinePublished in: ICML (2018)
Keyphrases
- reinforcement learning
- action selection
- partially observable domains
- reward shaping
- transition model
- state action
- action space
- reinforcement learning algorithms
- function approximation
- learning algorithm
- temporal difference
- sensory inputs
- state space
- fitted q iteration
- optimal control
- partially observable
- markov decision processes
- learning process
- action models
- optimal policy
- multi agent
- policy search
- computer vision
- function approximators
- temporal difference learning
- continuous state
- action descriptions
- robotic control
- machine learning
- continuous action