Solving a delayed response task with spiking and McCulloch-pitts agents

Abstract

This paper investigates the evolution of evolved autonomous agents that solve a memory-dependent delayed response task. Two types of neurocontrollers are evolved: networks of McCulloch-Pitts neurons, and spiky networks, evolving also the parameterization of the spiking dynamics. We show how the ability of a spiky neuron to accumulate voltage is utilized for the delayed response processing. We further confront new questions about the nature of "spikiness", showing that the presence of spiking dynamics does not necessarily transcribe to actual spikiness in the network, and identify two distinct properties of spiking dynamics in embedded agents. Our main result is that in tasks possessing memory-dependent dynamics, neurocontrollers with spiking neurons can be less complex and easier to evolve than neurocontrollers employing McCulloch-Pitts neurons. Additionally the combined utilization of spiking dynamics with incremental evolution can lead to the successful evolution of response behavior over very long delay periods.