On the convergence of projective-simulation-based reinforcement learning in Markov decision processes.
Walter L. BoyajianJens ClausenLea M. TrenkwalderVedran DunjkoHans J. BriegelPublished in: Quantum Mach. Intell. (2020)
Keyphrases
- markov decision processes
- reinforcement learning
- reinforcement learning algorithms
- optimal policy
- state space
- stochastic shortest path
- action space
- finite state
- reachability analysis
- policy iteration
- state and action spaces
- dynamic programming
- transition matrices
- convergence rate
- factored mdps
- reward function
- model based reinforcement learning
- stationary policies
- planning under uncertainty
- average reward
- finite horizon
- decision processes
- continuous state
- markov decision process
- state abstraction
- infinite horizon
- decentralized control
- decision theoretic planning
- control problems
- average cost
- policy iteration algorithm
- multi agent
- partially observable markov decision processes
- policy evaluation
- optimal control
- lead time
- real time dynamic programming