Q-Learning in enormous action spaces via amortized approximate maximization.
Tom Van de WieleDavid Warde-FarleyAndriy MnihVolodymyr MnihPublished in: CoRR (2020)
Keyphrases
- action space
- state space
- reinforcement learning
- action selection
- state action
- continuous state spaces
- reinforcement learning methods
- markov decision processes
- continuous state
- state and action spaces
- single agent
- function approximation
- reinforcement learning algorithms
- multi agent
- real valued
- skill learning
- dynamic programming
- control policies
- policy iteration
- function approximators
- model free
- dynamical systems
- optimal policy
- reinforcement learning problems
- state variables
- heuristic search
- objective function
- markov chain
- learning algorithm
- continuous action
- markov decision process
- stochastic processes
- planning problems
- multi agent systems
- temporal difference
- maximum entropy
- decision problems
- particle filter
- search space
- bayesian networks