Approximate Fisher Information Matrix to Characterize the Training of Deep Neural Networks.
Zhibin LiaoTom DrummondIan ReidGustavo CarneiroPublished in: IEEE Trans. Pattern Anal. Mach. Intell. (2020)
Keyphrases
- fisher information
- neural network
- training process
- positive definite
- information geometry
- probability density function
- euclidean space
- free energy
- noise level
- estimation error
- kullback leibler divergence
- exponential family
- geodesic distance
- projective transformations
- kl divergence
- multidimensional scaling
- competitive learning
- covariance matrix
- training samples
- edge detection
- training set