On Comparison between Evolutionary Programming Network-based Learning and Novel Evolution Strategy Algorithm-based Learning

TL;DR

This paper compares two evolutionary algorithms, EPNet and NES, for training neural networks, highlighting their different approaches to architecture evolution and training, and evaluates their performance on benchmark medical diagnosis problems.

Contribution

It introduces and compares EPNet and NES, novel evolutionary algorithms with distinct mechanisms for neural network training and architecture evolution.

Findings

EPNet evolves both network architecture and weights.

NES uses new genetic operators for training.

Both algorithms are tested on medical diagnosis benchmarks.

Abstract

This paper presents two different evolutionary systems - Evolutionary Programming Network (EPNet) and Novel Evolutions Strategy (NES) Algorithm. EPNet does both training and architecture evolution simultaneously, whereas NES does a fixed network and only trains the network. Five mutation operators proposed in EPNet to reflect the emphasis on evolving ANNs behaviors. Close behavioral links between parents and their offspring are maintained by various mutations, such as partial training and node splitting. On the other hand, NES uses two new genetic operators - subpopulation-based max-mean arithmetical crossover and time-variant mutation. The above-mentioned two algorithms have been tested on a number of benchmark problems, such as the medical diagnosis problems (breast cancer, diabetes, and heart disease). The results and the comparison between them are also presented in this paper.