A Novel Structured Natural Gradient Descent for Deep Learning

TL;DR

This paper introduces a new structured natural gradient descent method that reconstructs neural networks to approximate natural gradient optimization, improving convergence and performance while maintaining computational efficiency.

Contribution

It proposes reconstructing neural network structures to emulate natural gradient descent, offering a practical alternative that enhances training speed and accuracy.

Findings

Accelerates convergence of deep networks

Achieves better performance than traditional gradient descent

Maintains computational simplicity

Abstract

Natural gradient descent (NGD) provided deep insights and powerful tools to deep neural networks. However the computation of Fisher information matrix becomes more and more difficult as the network structure turns large and complex. This paper proposes a new optimization method whose main idea is to accurately replace the natural gradient optimization by reconstructing the network. More specifically, we reconstruct the structure of the deep neural network, and optimize the new network using traditional gradient descent (GD). The reconstructed network achieves the effect of the optimization way with natural gradient descent. Experimental results show that our optimization method can accelerate the convergence of deep network models and achieve better performance than GD while sharing its computational simplicity.