An Evolutionary Autonomous Agent with Visual Cortex and Recurrent Spiking Columnar Neural Network

Spiking neural networks are computationally more powerful than conventional artificial neural networks. Although this fact should make them especially desirable for use in evolutionary autonomous agent research, several factors have limited their application. This work demonstrates an evolutionary agent with a sizeable recurrent spiking neural network containing a biologically motivated columnar visual cortex. This model is instantiated in spiking neural network simulation software and challenged with a dynamic image recognition and memory task. Since the resulting agent or “virtual organism” initially contains many randomly and recurrently connected neurons, we use a genetic algorithm to evolve generations of this brain model that instinctively perform progressively better on the task. Our investigations lay the foundation for further experiments to resolve the question of whether autonomous agents with spiking neural networks can take advantage of the proven theoretical computat...

• Type: Conference
• Year: 2004
• Where: GECCO
• Authors: Rich Drewes, James B. Maciokas, Sushil J. Louis, Philip H. Goodman
• Keywords: Artificial Neural Networks, GECCO 2004, Sizeable Recurrent Spiking, Spiking Neural Networks