Gradient Descent Only Converges to Minimizers: Non-Isolated Critical Points and Invariant Regions.
Ioannis PanageasGeorgios PiliourasPublished in: ITCS (2017)
Keyphrases
- critical points
- saddle points
- scale space
- vector field
- feature points
- geometrical properties
- cost function
- scalar field
- gaussian scale space
- input image
- sufficient conditions
- affine transformation
- gaussian filter
- morse theory
- optimal solution
- objective function
- deep structure
- image sequences
- loss function
- facial expressions
- edge map
- saddle point
- computer vision