C++ Neural Networks and Fuzzy Logic: Preface

Algorithm for Kohonen’s Approach

A gain parameter » and a scale factor q are used while modifying the weights. A value between 0.02 and 0.2

was tried in previous examples for q. A distance of a neuron from the selected neuron is defined to be an

integer between 0 and n – 1, where n is the number of cities for the tour. This means that these distances are

not necessarily the actual distances between the cities. They could be made representative of the actual

distances in some way. One such attempt is described in the following discussion on C++ implementation.

This distance is denoted by d

j

 for neuron j. A squashing function similar to the Gaussian density function is

also used.

The details of the algorithm are in a paper referenced in Davalo. The steps of the algorithm to the extent given

by Davalo are:

•  Find the weight vector for which the distance from the input vector is the smallest

•  Modify the weights using

w

jnew

 = w

jold

 + (l

new

 − w

jold

)g(»,d

j

), where g(»,dj) = exp(− dj

2

/») /

•  Reset » as » (1 − q)

Previous Table of Contents Next

Copyright ©

 IDG Books Worldwide, Inc.

C++ Neural Networks and Fuzzy Logic:Preface

Other Approaches to Solve the Traveling Salesperson Problem

365

---