On the Variance of the Fisher Information for Deep Learning
arXiv:2107.04205
Abstract
In the realm of deep learning, the Fisher information matrix (FIM) gives novel insights and useful tools to characterize the loss landscape, perform second-order optimization, and build geometric learning theories. The exact FIM is either unavailable in closed form or too expensive to compute. In practice, it is almost always estimated based on empirical samples. We investigate two such estimators based on two equivalent representations of the FIM -- both unbiased and consistent. Their estimation quality is naturally gauged by their variance given in closed form. We analyze how the parametric structure of a deep neural network can affect the variance. The meaning of this variance measure and its upper bounds are then discussed in the context of deep learning.
arXiv commentPublished in Advances in Neural Information Processing Systems 34 (NeurIPS 2021)
References in corpus (9)
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Distilling the Knowledge in a Neural Network
- New insights and perspectives on the natural gradient method
- An elementary introduction to information geometry
- Hypothesis test for normal mixture models: The EM approach
- A New Generalization of Chebyshev Inequality for Random Vectors
- Limitations of the Empirical Fisher Approximation for Natural Gradient Descent
- Fisher Information and Natural Gradient Learning of Random Deep Networks
- Estimating Fisher Information Matrix in Latent Variable Models based on the Score Function