A Fisher-Rao gradient flow for entropy-regularised Markov decision processes in Polish spaces.
Bekzhan KerimkulovJames-Michael LeahyDavid SiskaLukasz SzpruchYufei ZhangPublished in: CoRR (2023)
Keyphrases
- markov decision processes
- gradient flow
- active contours
- energy functional
- partial differential equations
- finite state
- optimal policy
- level set
- active contour model
- transition matrices
- dynamic programming
- reinforcement learning
- curve evolution
- state space
- decision theoretic planning
- average cost
- policy iteration
- average reward
- model based reinforcement learning
- action space
- image dependent
- planning under uncertainty
- variational formulation
- reachability analysis
- energy minimization
- level set method
- markov decision process
- infinite horizon
- long run
- image restoration
- probabilistic model
- computer vision
- optimal control
- shape prior
- action sets
- linear programming
- image segmentation