B1956
Title: Higher-order connectivity network for multivariate point process data
Authors: Xiwei Tang - University of Virginia (United States) [presenting]
Abstract: High-dimensional point process modeling has emerged as a pivotal technique in the examination of neuronal spike trains. Unlike traditional point process models, which only acknowledge the additive effects of neurons within the calculated intensity for a target neuron, a groundbreaking model is introduced that embraces the higher-order interactions among neurons by employing a multivariate convolution. The pertinent transferring coefficients are organized in a three-way tensor format, and we subsequently enforce low-rank, sparsity, and subgroup structures upon this coefficient tensor. These imposed structures facilitate dimensionality reduction, enable the synthesis of information across disparate individual processes, and augment interpretative ease. A highly scalable optimization algorithm is developed for precise parameter estimation, accompanied by the establishment of theoretical guarantees for both the algorithm and large-sample properties.