Convergent evolution of neuron architectures

A universal structural pattern of cognitive architectures is the presence of a class of fundamental units that display polarity (information is sent in one direction) and threshold response dynamics. Despite the potential diversity of design principles, this architecture suggests a convergent solution to the problem of sensing and processing information in natural systems. In this work, we explore this possibility by evolving artificial systems of connected units (cells) that start from a homogeneous network of linear elements, with the capability of developing a non-linear activation function.