Tabu and Scatter Search for Artificial Neural Networks

Abstract

In this paper we address the problem of training multilayer feed-forward neural networks. These networks have been widely used for both prediction and classification in many different areas. Although the most popular method for training these networks is back propagation, other optimization methods such as tabu search or scatter search have been applied to solve this problem. This paper presents a new training algorithm based on the tabu search methodology that incorporates elements for search intensification and diversification by utilizing strategic designs where other previous approaches resort to randomization. Our method considers context and search information, as it is provided by the partial derivatives and memory structures, for move selection. The experimentation shows that the proposed procedure can compete with the best-known algorithms in terms of solution quality, consuming a reasonable computational effort.