The focus of this paper is the examination of the temporal pattern recognition capacities of a simplified action potential model similar to the one proposed by Hopfield et al. The features of a simplified model that differ from the Hopfield model are the lack of logarithmic scaling, addition of speed to set the pattern, and a more thorough specification of what is meant, by synchronicity, by explicitly identifying an activation window where approximate alignment causes output activation. The simplified model is shown to work with the Analogue Match problem. Two further extensions of the Analogue Match problem are proposed. The first incorporates noise into the signal intensity (noisy inputs) and the second introduces noise into the recognition vector (noisy patterns). The simplified action potential model is shown to function in the short-term using a fixed window so long as the noise is in proportion to the signal intensity.
ACM International Conference on Computing Frontiers 2015, Ischia, Italy, 18-21 May 2015