Geometry of Neural Network Loss Surfaces via Random Matrix Theory.
Jeffrey PenningtonYasaman BahriPublished in: ICML (2017)
Keyphrases
- neural network
- random matrix theory
- three dimensional
- surface matching
- correlation matrix
- surface geometry
- geometric modeling
- surface representation
- pattern recognition
- geometric features
- free form
- neural network model
- back propagation
- reflectance properties
- artificial neural networks
- geometric models
- photometric properties
- associative memory
- surface reconstruction
- high dimensional
- geometric primitives
- smooth surfaces
- d objects
- support vector
- computer vision
- photometric information
- machine learning