Evolving spiking neural networks for temporal pattern recognition in the presence of noise

Nervous systems of biological organisms use temporal patterns of spikes to encode sensory input, but the mechanisms that underlie the recognition of such patterns are unclear. In the present work, we explore how networks of spiking neurons can be evolved to recognize temporal input patterns without being able to adjust signal conduction delays. We evolve the networks with GReaNs, an artificial life platform that encodes the topology of the network (and the weights of connections) in a fashion inspired by the encoding of gene regulatory networks in biological genomes. The number of computational nodes or connections is not limited in GReaNs, but here we limit the size of the networks to analyze the functioning of the networks and the effect of network size on the evolvability of robustness to noise. Our results show that even very small networks of spiking neurons can perform temporal pattern recognition in the presence of input noise