Dynamic Neural Diversification: Path to Computationally Sustainable Neural Networks

TL;DR

This paper investigates how increasing neuron diversity within small neural networks during training can improve learning efficiency and accuracy, proposing techniques for dynamic decorrelation and a novel weight initialization method.

Contribution

It introduces methods for dynamically enhancing neuron diversity during training and a new weight initialization technique that accelerates learning in small neural networks.

Findings

Decorrelated weight initialization yields 40% relative accuracy increase in early epochs.

Diversity reinforcement techniques improve early learning and help escape local minima.

Methods are effective for resource-efficient neural network training.

Abstract

Small neural networks with a constrained number of trainable parameters, can be suitable resource-efficient candidates for many simple tasks, where now excessively large models are used. However, such models face several problems during the learning process, mainly due to the redundancy of the individual neurons, which results in sub-optimal accuracy or the need for additional training steps. Here, we explore the diversity of the neurons within the hidden layer during the learning process, and analyze how the diversity of the neurons affects predictions of the model. As following, we introduce several techniques to dynamically reinforce diversity between neurons during the training. These decorrelation techniques improve learning at early stages and occasionally help to overcome local minima faster. Additionally, we describe novel weight initialization method to obtain decorrelated, yet stochastic weight initialization for a fast and efficient neural network training. Decorrelated weight initialization in our case shows about 40% relative increase in test accuracy during the first 5 epochs.