伟德体育学术报告[2026]072号

（高水平大学建设系列报告1331号）

报告题目：Causal inference for multivariate point process treatments via instrumental variables

报告人：陈施喆 副教授（美国加州大学戴维斯分校）

报告时间：2026年7月1日11:00—12:00

报告地点：Bevictor伟德官网粤海校区汇星楼514会议室

报告摘要：Estimating causal effects from multivariate point process data is a fundamental challenge in many scientific fields. A prominent example arises in neuroscience, where spike trains are recorded simultaneously across multiple brain areas to study how activity in some areas causally influences spiking dynamics in others. In these studies, rigorous causal inference is complicated by unobserved confounding arising from unrecorded neural activities. We propose an instrumental variable method for settings with multivariate point process treatments. We establish conditions under which the causal effects are nonparametrically identified as the unique solution to a system of multivariate convolution equations. Based on this characterization, we develop a penalized estimator and establish its consistency and asymptotic normality.  We further propose a robust hypothesis testing procedure based on a Karhunen-Loeve expansion. We apply the proposed method to Neuropixels recordings from a visual discrimination task in mice, demonstrating how activity in the visual cortex and thalamic nuclei causally influences firing patterns in the hippocampal formation.

报告人简介：陈施喆博士现任美国加州大学戴维斯分校统计系副教授。此前，他曾在哥伦比亚大学统计系及 Grossman Center for the Statistics of Mind 从事博士后研究；他于华盛顿大学获得生物统计学博士学位。他的研究兴趣主要集中在高维统计、图模型、点过程、网络推断、因果推断及其在神经科学中的应用。他的研究关注如何从大规模复杂数据中学习生物系统和神经系统的结构与动态，并发展具有理论保证和实际可解释性的统计方法。

邀请人：张一弛

伟德体育

2026年6月29日