Learning Deep Latent-variable MRFs with Amortized Bethe Free Energy Minimization.
Sam WisemanPublished in: DGS@ICLR (2019)
Keyphrases
- max flow min cut
- energy minimization
- markov random field
- graph cuts
- energy function
- latent variables
- latent variable models
- image segmentation
- low level vision
- reinforcement learning
- belief propagation
- higher order
- machine learning
- learning problems
- weighted constraint satisfaction
- message passing
- dual decomposition
- conditional random fields
- free energy
- maximum likelihood
- knowledge discovery
- prior knowledge
- learning algorithm