Title: Point singularities and suprathreshold stochastic resonance in optimal coding

Abstract: Motivated by recent studies of population coding in theoretical neuroscience, we examine the optimality of a recently described form of stochastic resonance known as suprathreshold stochastic resonance, which occurs in populations of noisy threshold devices such as models of sensory neurons. Using the mutual information measure, it is shown numerically that for a random input signal, the optimal threshold distribution contains singularities. For large enough noise, this distribution consists of a single point and hence the optimal encoding is realized by the suprathreshold stochastic resonance effect. Furthermore, it is shown that a bifurcational pattern appears in the optimal threshold settings as the noise intensity increases. Fisher information is used to examine the behavior of the optimal threshold distribution as the population size approaches infinity.