Generalization bound of globally optimal non-convex neural network training: Transportation map estimation by infinite dimensional Langevin dynamics.
Taiji SuzukiPublished in: NeurIPS (2020)
Keyphrases
- globally optimal
- infinite dimensional
- map estimation
- neural network training
- maximum a posteriori
- finite dimensional
- graph cuts
- markov random field
- neural network
- maximum likelihood
- training algorithm
- energy function
- convex sets
- em algorithm
- optimization method
- noisy images
- expectation maximization
- higher dimensional
- shape space
- image reconstruction
- closed curves
- belief propagation
- euclidean space
- shape model
- image restoration
- image registration
- feature vectors
- image segmentation