---
title: "Avoiding Overfitting with BP-SOM"
authors:
  - "Ton Weijters"
  - "H. Jaap van den Herik"
  - "Antal van den Bosch"
  - "Eric O. Postma"
doi: ""
url: "https://ijcai.org/proceedings/1997/165"
pdf: "https://ijcai.org/proceedings/1997/0165.pdf"
venue: "Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence"
year: 1997
pages: "1140-1144"
track: "Neural Networks"
keywords:
---

# Avoiding Overfitting with BP-SOM

Ton Weijters, H. Jaap van den Herik, Antal van den Bosch, Eric O. Postma

*Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence*, pages 1140-1144.

## Abstract

Overfitting is a well-known problem in the fields of symbolic and connectionist machine l ea rning. It describes the deterioration of generalisation performance of a trained model. In this paper , we investigate the ability of a novel artificial neural network , BP -S OM , to avoid overfitting. BP-SOM is a hybrid neural network which combines a multi-layered feed-forwa rd network (MFN) with Kohon en 's self-organising maps (soMs). During training, supervised back-propagation learning and unsupervised SOM learning cooperate in finding adequate hidden-layer representations . We show that BP-SOM outperforms standard backpropagation, and also back-propagation with a weight decay when dealing with the problem of overfitting. In addition, we show that BP-SOM succeeds in preserving generalisation performance under hidden-unit pruning , where both other m et hods fail.