Using bisimulation for policy transfer in MDPs.
Pablo Samuel CastroDoina PrecupPublished in: AAMAS (2010)
Keyphrases
- optimal policy
- markov decision processes
- markov decision problems
- markov decision process
- policy search
- finite horizon
- policy iteration
- reinforcement learning
- state and action spaces
- average cost
- average reward
- partially observable
- reward function
- state space
- action space
- infinite horizon
- policy evaluation
- dynamic programming
- control policies
- transfer learning
- factored mdps
- finite state
- long run
- factored markov decision processes
- knowledge transfer
- partially observable markov decision processes
- decision processes
- discounted reward
- state dependent
- transition probabilities
- linear programming
- model free
- reinforcement learning problems
- approximate policy iteration
- semi markov decision processes
- policy gradient
- inverse reinforcement learning
- decision theoretic planning
- continuous state spaces
- approximate dynamic programming
- finite number
- control policy
- equivalence relation
- action selection
- initial state