Deep Reinforcement Learning with Attention for Slate Markov Decision Processes with High-Dimensional States and Actions.
Peter SunehagRichard EvansGabriel Dulac-ArnoldYori ZwolsDaniel VisentinBen CoppinPublished in: CoRR (2015)
Keyphrases
- continuous state spaces
- markov decision processes
- action space
- reinforcement learning
- perceptual aliasing
- state action
- state space
- state and action spaces
- initial state
- partially observable
- continuous state
- markov decision problems
- optimal policy
- reward function
- action sets
- decision theoretic planning
- state abstraction
- decision processes
- average reward
- reinforcement learning algorithms
- real time dynamic programming
- policy iteration
- finite state
- transition matrices
- state transition
- dynamic programming
- markov decision process
- state transitions
- planning under uncertainty
- model based reinforcement learning
- macro actions
- reachability analysis
- model free
- function approximators
- factored mdps
- partially observable markov decision process
- rl algorithms
- action selection
- infinite horizon
- function approximation
- stochastic games
- partially observable markov decision processes
- transition model
- average cost
- state variables
- situation calculus
- discounted reward
- multi agent
- decentralized control
- real valued
- decision problems
- sufficient conditions