Deep Recurrent Q-Learning for Partially Observable MDPs.
Matthew J. HausknechtPeter StonePublished in: CoRR (2015)
Keyphrases
- partially observable
- reinforcement learning
- state space
- markov decision processes
- discount factor
- reward function
- reinforcement learning algorithms
- markov decision problems
- optimal policy
- policy iteration
- partial observability
- stochastic shortest path
- decision problems
- infinite horizon
- dynamical systems
- model free
- dynamic programming
- continuous state spaces
- partial observations
- heuristic search
- function approximation
- continuous state and action spaces
- partially observable environments
- action models
- finite state
- orders of magnitude
- markov chain
- average cost
- action selection
- multi agent
- learning algorithm
- dec pomdps
- partially observable markov decision process
- probabilistic planning
- machine learning
- special case
- state action
- average reward
- action space
- learning agent
- multiple agents
- state variables
- decision processes
- fully observable
- markov decision process
- generative model
- long run