Evolution of Robust Developmental Neural Networks

Abstract

We present the first evolved solutions to a computational task within the Neuronal Organism Evolution model (Norgev) of artificial neural network development. These networks display a remarkable robustness to external noise sources, and can regrow to functionality when severely damaged. In this framework, we evolved a doubling of network functionality (double-NAND circuit). The network structure of these evolved solutions does not follow the logic of human coding, and instead more resembles the decentralized dendritic connection pattern of more biological networks such as the 'C. elegans' brain.