Supplemental Material for Functional Identification of Spike-Processing Neural Circuits

In order to properly compare identification results between our methodology and the GLM framework, we used the same filter kernels and fixed the input signals used in identification. We also fixed the basis functions with respect to which both methods reconstruct the kernels. The latter was needed since the GLM typically employs basis functions that favor very particular kernels: those that oscillate rapidly close to the origin and have a very coarse structure further away from the origin (although the method can work with any adequately chosen basis (Pillow et al., 2008)). These kernels are shown in Fig. S1. While such an assumption about the filters might hold for some neural circuits, it is best to use a basis that can represent an arbitrary function on a given time