Fig. 1

A schematic representation of the averaging algorithm. A is a raster of spiking responses to a single stimulus. These are converted to functions by filtering (a). B shows the collection of functions that results. These are averaged (b) giving the average function C. This average function is approximated (c) by D, a function which itself is the result of filtering. This filtering is represented by d and the corresponding spike train by E. The optimal choice of E is the central spike train $\overline{\mathbf{u}}$. The greedy algorithm is used to estimate this