Adaptive multiple optimal learning factors for neural network training

TL;DR

This paper introduces the AMOLF algorithm that adaptively adjusts learning factors during neural network training, improving efficiency and accuracy by dynamically responding to error changes.

Contribution

It presents a novel adaptive algorithm for neural network training that optimizes learning factors and includes techniques for weight grouping and Hessian matrix compression.

Findings

AMOLF outperforms existing methods like OWO-MOLF and Levenberg-Marquardt.

Dynamic adjustment of learning factors enhances training efficiency.

Weight grouping and Hessian compression improve computational performance.

Abstract

This thesis presents a novel approach to neural network training that addresses the challenge of determining the optimal number of learning factors. The proposed Adaptive Multiple Optimal Learning Factors (AMOLF) algorithm dynamically adjusts the number of learning factors based on the error change per multiply, leading to improved training efficiency and accuracy. The thesis also introduces techniques for grouping weights based on the curvature of the objective function and for compressing large Hessian matrices. Experimental results demonstrate the superior performance of AMOLF compared to existing methods like OWO-MOLF and Levenberg-Marquardt.