New Algorithms for Encoding, Learning and Classification of fMRI Data in a Spiking Neural Network Architecture: A Case on Modeling and Understanding of Dynamic Cognitive Processes

Abstract

This paper argues that, the third generation of neural networks—the spiking neural networks (SNNs), can be used to model dynamic, spatio-temporal, cognitive brain processes measured as functional magnetic resonance imaging (fMRI) data. This paper proposes a novel method based on the NeuCube SNN architecture for which the following new algorithms are introduced: fMRI data encoding into spike sequences; deep unsupervised learning of fMRI data in a 3-D SNN reservoir; classification of cognitive states; and connectivity visualization and analysis for the purpose of understanding cognitive dynamics. The method is illustrated on two case studies of cognitive data modeling from a benchmark fMRI data set of seeing a picture versus reading a sentence.