Pathological spectra of the Fisher information metric and its variants in deep neural networks.
Ryo KarakidaShotaro AkahoShun-ichi AmariPublished in: CoRR (2019)
Keyphrases
- fisher information
- neural network
- information geometry
- riemannian metric
- pattern recognition
- probability density function
- euclidean space
- noise level
- positive definite
- free energy
- multidimensional scaling
- competitive learning
- kullback leibler divergence
- kl divergence
- exponential family
- nearest neighbor
- support vector machine
- feature selection