2009 Special Issue: Error-probability noise benefits in threshold neural signal detection

摘要: Five new theorems and a stochastic learning algorithm show that noise can benefit threshold neural signal detection by reducing the probability of error. The first theorem gives necessary sufficient condition for such when neuron performs discrete binary in presence additive scale-family noise. allows user to find optimal density several closed-form types include generalized Gaussian second noise-benefit more general signals have continuous densities. third fourth reduce this weighted-derivative comparison densities at are continuously differentiable is symmetric comes from scale family. fifth shows how collective benefits occur parallel array neurons even an individual does not itself produce benefit. gradient-ascent value do closed form.