Lagrange Dual Decomposition for Finite Horizon Markov Decision Processes.
Thomas FurmstonDavid BarberPublished in: ECML/PKDD (1) (2011)
Keyphrases
- dual decomposition
- finite horizon
- markov decision processes
- optimal policy
- energy minimization
- map inference
- infinite horizon
- state space
- lagrangian relaxation
- finite state
- average cost
- dynamic programming
- markov random field
- markov decision process
- reinforcement learning
- markov logic
- action space
- max margin
- graphical models
- energy function
- machine learning
- belief propagation
- long run
- sufficient conditions
- reward function
- lower bound