ICML-98 Submission #84

An Experimental Evaluation of Coevolutive Concept Learning

C. Anglano, A. Giordana, G. Lo Bello, L. Saitta
Dipartimento di Informatica, Universita` di Torino
corso Svizzera 185, 10149 Torino Italy

Abstract

In this paper an extensive experimental evaluation of an evolutionary 
approach to concept learning is presented. The experimentation, 
performed with the system G-NET, investigates the effectiveness of the 
approach along the following dimensions: robustness with respect to 
parameter setting, effectiveness of the MDL criterion coupled with a 
stochastic search bias, impact of coevolution on the quality of the 
solution and on the computational effort required, and ability to face 
problems requiring structured representation languages. A discussion of 
the obtained results and a suggestion on when this type of 
approach might be useful is also provided.

Keywords: Inductive Learning, Evolutionary Computation, MDL Principle.

Contact Author: Attilio Giordana

E-mail Address of Contact Author: attilio@di.unito.it

Phone Number of Contact Author: +39-11-7429215