Automatic discovery genetic ii program programming reusable

Riolo eds. Falkenhainer and R. Kodratoff, and R. Michalski eds. Heal, M. Hansen, and K. Hunt, R. Lipsman, and J. Iba, H. Kinnear, Jr. Keane, J. Koza, and J. Koza, D. Andre, F. Bennett III, and M. Langley, G. Bradshaw, and H. Langley and J. Spector, W. Langdon, U. O'Reilly, and P. Angeline eds. Luke and L. Koza, K. Garzon, H. Moustafa, K. De Jong, and E. Banzhaf, J. Daida, A. Eiben, M.

Garzon, V. Honavar, M. Jakiela, and R. Smith, eds. Ryan, J. Collins, and M. Banzhaf, R. Press, A Bradford Book Cloth: ISBN Order this book This book was published two years after the first collection, when the field of GP had settled down more into what might be called 'normal science'.

It is thus far more concerned with technical issues and is generally of far less interest to the social simulator. What it does do is start to examine and compare the various varieties of GP in a more methodical manner. Thus the various chapters do provide the reader with many ideas for refining and improving the efficiency of GP in different domains.

However, being merelya collection of papers, it does not do this in a systematic manner, as a handbook might. Three chapters are worth mentioning. Genetic Programming II extends the results of John Koza's ground-breaking work on programming by means of natural selection, described in his first book, Genetic Programming. Using a hierarchical approach, Koza shows that complex problems can be solved by breaking them down into smaller, simpler problems using the recently developed technique of automatic function definition in the context of genetic programming.

Where conventional techniques of machine learning and artificial intelligence fail to provide an effective means for automatically handling the process of decomposing complex problems into smaller subsets, reassembling the solutions to these subsets, and applying an overall solution to the original problem, automatic function definition enables genetic programming to define useful and reusable subroutines dynamically.

Koza illustrates this new technique by showing how it solves or approximately solves a variety of problems in Boolean function learning, symbolic regression, control, pattern recognition, robotics, classification, and molecular biology. In each example, the problem is automatically decomposed into subproblems; the subproblems are automatically solved; and the solutions to the subproblems are automatically assembled into a solution to the original problem.

Koza shows that leverage accrues because genetic programming with automatic function definition repeatedly uses the solutions to the subproblems in the assembly of the solution to the overall problem.