Discovery of Salient Low-Dimensional Dynamical Structure in Neuronal Population Activity Using Hopfield Networks


September 19, 2016 - 2:00pm - 3:15pm
Northwest 243
About the Speaker
Dr. Felix Effenberger

We present a novel method for the classical task of finding and extracting recurring spatiotemporal patterns in parallel spike trains. In contrast to previously proposed methods it does not suffer from combinatorial explosion and does not seek to classify exactly recurring patterns, but rather approximate versions possibly differing by a certain number of missed, shifted or excess spikes. We achieve this by fitting high-dimensional Ising models to windowed, binned spiking activity in an unsupervised way using minimum probability flow parameter estimation and then collect memories of the associated Hopfield network over the raw data. By modeling the sequence of occurring memories over the original data as a Markov process, we are able to extract low-dimensional representations of neural population activity on longer time scales.

We demonstrate the method on a data set obtained in rat barrel cortex and show that it is able to extract a remarkably low-dimensional, yet accurate representation of mean population activity observed during the experiment. The presented method is furthermore applicable in greater generality to the unsupervised de-noising of high-dimensional noisy data as will be demonstrated for the instance of images obtained by serial electron microscopy. An implementation is available in form of the freely available open source Python package hdnet. This is joint work with Christopher Hillar, MSRI / Redwood Center / UCSF.