Pathological Spectra of the Fisher Information Metric and Its Variants in Deep Neural Networks.
Ryo KarakidaShotaro AkahoShun-ichi AmariPublished in: Neural Comput. (2021)
Keyphrases
- fisher information
- neural network
- probability density function
- information geometry
- euclidean space
- pattern recognition
- noise level
- positive definite
- free energy
- riemannian metric
- geodesic distance
- self organizing maps
- metric space
- competitive learning
- kullback leibler divergence
- multidimensional scaling
- exponential family
- kl divergence
- error rate
- principal component analysis