Detecting and measuring higher order synchronization among neurons: A Bayesian Approach

Abstract

A Bayesian approach to modeling and inferring patterns of synchronous activation in a group of neurons. A major objective of the research is to provide statistical tools for detecting changes in synchronization patterns. Our framework is not restricted to the case of correlated pairs, but generalizes the Boltzmann machine model to allow for higher order interactions. A Markov Chain Monte Carlo Model Composition (MC3) algorithm is applied in order to search over connectivity structures and uses Laplace's method to approximate their posterior probabilities.Performance of the method was first tested on synthetic data. The method was then applied to data obtained on multi-unit recordings of six neurons in the visual cortex of a rhesus monkey in two different attentional states. The obtained results indicate that the interaction structure predicted by the data is richer than just a set of synchronous pairs. They also confirmed the experimenter's conjecture that different attentional states were associated with different interaction structures.