Architecture Optimization and Training for the Multilayer Perceptron using Ant System

摘要

We present in this paper an Ant Colony Algorithm to optimize the performance of the multilayer Perceptron. Indeed, the performance of the multilayer Perception depends on its parameters such as the number of neurons in the hidden layer and the connection weights. In this respect, we firstly model the problem of neural architecture and training in terms of a mixed-integer problem with a linear constraint, and secondly, we propose an Ant Colony Algorithm to solve it. The experimental results illustrate the advantage of our approach as a new method of training and architecture optimization.