3 Koza, John R. Andre, David 1996 Lawrence Hunter and Teri E. Klein Pacific Symposium on Biocomputing '96 World Scientific 500--511 genetic algorithms, genetic programming This paper describes how the biological theory of gene duplication described in Susumu Ohno's provocative book, Evolution by Means of Gene Duplication, was brought to bear on a vexatious problem from the domain of automated machine learning, namely the problem of architecture discovery. An automatic programming system should require that the user make as few problem-specific decisions as possible concerning the size, shape, and character of the ultimate solution to the problem. Six new architecture-altering operations enable genetic programming to automatically discover an appropriate architecture for solving the problem concurrently with its efforts to solve the problem. These architecture-altering operations were motivated by the way that new biological structures, functions, and behaviors arise in nature using gene duplication. Genetic programming with the new architecture-altering operations was then applied to the transmembrane protein segment identification problem. The out-of-sample error rate for the best genetically-evolved program achieved was slightly better than that of previously-reported human-written algorithms for this problem. http://www.genetic-programming.com/jkpdf/psb1996.pdf