New methods for simulation and analysis of correlated spike-trains

As signals propagate through nonlinear systems their auto- and cross-correlation functions are distorted. This distortion can be analytically solved for Gaussian signals undergoing several common nonnegative transformations. We show how this solution, together with linear modeling techniques, is useful both for flexible generation of synthetic spike trains with pre-defined auto- and cross-correlation functions, and, conversely, for the identification of Linear-Nonlinear-Poisson (LNP) encoding models purely from the given systems input and output autocorrelation structures. Such correlation-based identification is a dasiablindpsila alternative to reverse correlation and related techniques.