Deterministic policies based on maximum regrets in MDPs with imprecise rewards.
Pegah AlizadehEmiliano TraversiAomar OsmaniPublished in: AI Commun. (2021)
Keyphrases
- markov decision processes
- reward function
- optimal policy
- stationary policies
- fully observable
- markov decision problems
- reinforcement learning
- markov decision process
- total reward
- expected reward
- discounted reward
- state space
- decision processes
- policy iteration
- finite state
- partially observable markov decision processes
- average cost
- partially observable
- average reward
- control policies
- decision problems
- reinforcement learning algorithms
- action sets
- policy search
- linear programming
- decentralized control
- factored mdps
- dynamic programming
- finite horizon
- planning under uncertainty
- action space
- decision theoretic planning
- partial observability
- decision diagrams
- state and action spaces
- multistage
- multiple agents
- initial state
- expected utility
- stochastic domains
- infinite horizon
- long run
- decision theoretic
- factored markov decision processes