An additive noise model followed by a nonlinearity explains a diversity of neural over-dispersion

What is it about?

We model here how a neuron's spike rate as containing additive gaussian noise that is transformed nonlinearly to produce a Poisson spike rate. Different nonlinearities give rise to qualitatively different mean-variance relationships, ranging from sublinear to linear to quadratic. We describe a computationally efficient method for fitting this model to data and demonstrate that a majority of neurons in a V1 population are better described by a model with a nonquadratic mean-variance relationship. Finally, we demonstrate a practical use of our model via an application to Bayesian adaptive stimulus selection in closed-loop neurophysiology experiments, which shows that accounting for overdispersion can lead to dramatic improvements in adaptive tuning curve estimation.

Featured Image

Why is it important?

Understanding how to model neural spiking more accurately allows for the disentanglement of the task- related portion of a neuron's spiking rate from the overall rate. This disentanglement can lead, as is shown in the paper, to superior rate estimation that is useful in tasks such as creating closed-loop experiments.