Modular neural networks evolved by genetic programming
Sung‐Bae Cho, Katsunori Shimohara · 2002
We present an evolvable model of modular neural networks which are rich in autonomy and creativity. In order to build an artificial neural network which is rich in autonomy and creativity, we have adopted the ideas and methodologies of artificial life. The paper describes the concepts and methodologies for the evolvable model of modular neural networks, which will be able not only to develop new functionality spontaneously but also to grow and evolve its own structure autonomously. Although the ultimate goal of this model is to design the control system for such behavior based robots as Khepera, we have attempted to apply the mechanism to a visual categorization task with handwritten digits. The evolutionary mechanism has shown a strong possibility to generate useful network architectures from an initial set of randomly connected networks.