Mathematics Colloquia and Seminars

Return to Colloquia & Seminar listing

Multivariate point processes are popular tools for inferring relationships among subjects from recurrent event data such as neural spike trains. Complicated by the unmeasured confounding variables, interventions to the system are often employed in order to infer causality. However, these interventions are of low precision that they might influence the intensities of multiple processes simultaneously. In this study, we propose an instrumental variable framework with treatments being multivariate point processes. We show that the causal effects can be learned using generalized Wald estimation. We propose a penalized estimation procedure motivated by classic methods for density deconvolution. The proposed method is applied to neural data from behavioral experiments on mice. This is joint work with Yu Liu and Zhichao Jiang.