Next: Introduction

Better Trained Ants

W. B. Langdon
School of Computer Science, The University of Birmingham, Birmingham B15 2TT, UK
W.B.Langdon@cs.bham.ac.uk http://www.cs.bham.ac.uk/wbl
Tel: +44 (0) 121 414 4791, Fax: +44 (0) 121 414 4281

Technical Report: CSRP-98-08
28 February 1998

Abstract:

The problem of programming an artificial ant to follow the Santa Fe trail has been repeatedly used as a benchmark problem. Recently we have shown performance of several techniques is not much better than the best performance obtainable using uniform random search. We suggested that this could be because the program fitness landscape is difficult for hill climbers and the problem is also difficult for Genetic Algorithms as it contains multiple levels of deception.

Here we redefine the problem so the ant is obliged to traverse the trail in approximately the correct order. A simple genetic programming system, with no size or depth restriction, is show to perform approximately three times better with the improved training function.

William B Langdon
Wed Mar 4 13:21:44 GMT 1998