Pattern recognition with spiking neurons: performance enhancement based on a statistical analysis

Abstract

1 2 29 3, (godin@bruyeres.cea.fr), J. D. Muller ( l ) (muller@ldg.bruyeres.cea.fi), M. B. Gordon (2)(gordon@drfmc.ceng.cea.fi) and J. Haussy (')(hausy@bruyeres.cea.fi) Abstract PCIW (Pulse Coupled Neural Networh) and more generally spiking-neuron models seem to meet the realtime and robustness constraints necessav in on-board pattern recognition applications. However, ejicient learning algorithms are still lacking for such networks. In this paper we consider a feedgoward network of spiking neurons. m e weights and bioses are obtained after a simple transfonnation of those learned with standard back-propagation on a static (stanah-4 neural network We discuss the conditions under which this transformationgives good recognition rates, in the case of handwritten digit recognition.

Review of pulse-coupled neural networks

“…Godin et al enhance the performance of recognition via introducing statistical analysis into the PCNN model and gain the higher recognizing rate in the case of handwritten digit recognition [78]. Allen et al propose a way of how the PCNN bridges the gap between syntactic and statistical pattern recognition [79].…”