Approximate Fisher Information Matrix to Characterise the Training of Deep Neural Networks

TL;DR

This paper introduces a novel method using eigenvalues of an approximate Fisher information matrix to monitor and optimize deep neural network training, improving convergence and generalisation.

Contribution

It presents a new measurement technique based on Fisher information eigenvalues and a dynamic sampling training approach for better training efficiency and accuracy.

Findings

Effective monitoring of training via Fisher eigenvalues

Dynamic sampling improves training speed and accuracy

Method applicable to high-capacity deep models

Abstract

In this paper, we introduce a novel methodology for characterising the performance of deep learning networks (ResNets and DenseNet) with respect to training convergence and generalisation as a function of mini-batch size and learning rate for image classification. This methodology is based on novel measurements derived from the eigenvalues of the approximate Fisher information matrix, which can be efficiently computed even for high capacity deep models. Our proposed measurements can help practitioners to monitor and control the training process (by actively tuning the mini-batch size and learning rate) to allow for good training convergence and generalisation. Furthermore, the proposed measurements also allow us to show that it is possible to optimise the training process with a new dynamic sampling training approach that continuously and automatically change the mini-batch size and learning rate during the training process. Finally, we show that the proposed dynamic sampling training approach has a faster training time and a competitive classification accuracy compared to the current state of the art.