A Massively Parallel Approach for Nonlinear Interdependency Analysis of Multivariate Signals with GPGPU

Abstract

Nonlinear interdependency (NLI) analysis is an effective method for measurement of synchronization among brain regions, which is an important feature of normal and abnormal brain functions. But its application in practice has long been largely hampered by the ultra-high complexity of the NLI algorithms. We developed a massively parallel approach to address this problem. The approach has dramatically improved the runtime performance. It also enabled NLI analysis on multivariate signals which was previously impossible.