Summary


The paper deals with a description of a constructive neural network based on gradient initial setting of its weights. The network has been used as a pattern classifier of two dimensional patterns but it can be generally used to n × m associative problems. A network topology, processes of learning and retrieving, experiments and comparison to other neural networks are described.

See the full content of this document

Extract


Constructive Gradient Neural Network

1. Introduction

Multilayer networks with fixed topology are the most commonly used artificial neural network models. However, the topology of such a network must be specified before the learning process of the network. The only way to specify the topology is often an "ad hoc" approach which depends on personal experience.

Therefore the algorithms that can find sub-optimal neural network topology are in the focus of the research in this field. The minimal topology able to solve a given problem is considered as the optimal one. There are two basic approaches. Pruning algorithms start with learning of a network big enough to solve the given problem and then some redundant weights or neurons are pruned out of the network. A constructive approach starts with a small network (often an elementary one) and adds new neurons and weights together with their learning until the problem is solved...

See the full content of this document

Sponsored links




ver las páginas en versión mobile | web

ver las páginas en versión mobile | web

© Copyright 2012, vLex. All Rights Reserved.

Contents in vLex Germany

Explore vLex

For Professionals

For Partners

Company