Attractor-Based Pattern Classification in a Spiking FPGA Implementation of the Olfactory Bulb

A programmable logic implementation of a spiking neuronal olfactory bulb is presented, based upon a construction kit of leaky integrate-and-fire soma and dynamical synapse models that achieve exact integration performance operating at least 4 orders of magnitude faster than biological timescales. We show how neuronal elements from this construction kit may be deployed in parallel, following the known architecture of the early olfactory pathway, which when combined spontaneously generates oscillatory, limit-cycle, dynamics reminiscent of those found in biology. When incorporated with a correlation learning rule, in the form of Hebbian learning, we demonstrate that the system is able to store attractors which correspond to learnt input odor patterns. Our results show how these attractors may be read-out from the network in order to classify previously encountered learnt odors without confusion. Such architectures may be useful in solving particularly challenging odor identification tasks such as segmentation of complex odors or identification of complex odors within an interfering and nonstationary background