Modular NEAT and Continued Work on ecGA

by J.D. Marble on February 17, 2010

Overview

  • Modular NeuroEvolution of Augmenting Topologies (Modular NEAT) [1]

  • OneMax vs. Deceptive problems with cGA

Modular NEAT

Graphical indication of how a module is bound into the solution network.[1]

OneMax with cGA

Performance of cGA on various sized problems with pop=100

Performance of cGA on various sized problems with pop=100

OneMax with cGA

Performance of cGA on various sized problems with pop=50

Performance of cGA on various sized problems with pop=50

OneMax with cGA

Performance of cGA on various sized problems with pop=10

Performance of cGA on various sized problems with pop=10

Deceptive Problem

Description of order–4 deceptive problem[2]

Bibliography

[1] J. Reisinger, K.O. Stanley, and R. Miikkulainen, “Evolving Reusable Neural Modules,” Genetic and Evolutionary Computation – GECCO 2004, 2004, pp. 69–81.

[2] G. Harik, “Linkage learning via probabilistic modeling in the ECGA,” Urbana, IL: University of Illinois at Urbana-Champaign, 1999.