Beyond GLMs: A Generative Mixture Modeling Approach to Neural System Identification
Figure 1
Illustration of the spike-triggered mixture model (STM).
A. A sigmoidal nonlinearity is applied to a log-likelihood ratio of two mixtures of Gaussians to determine the firing rate of the model, which is then used to generate spikes. B. By making a naive Bayes assumption, additional information and measurements such as interspike interval distributions can easily be incorporated into the model in the form of additional log-likelihood ratios.