革新知能統合研究センター

Speaker: Hongyi Ding (The University of Tokyo)

Title: Variational Bayesian Inference of Point Processes for Time-Sequence Modeling

Abstract:
A time-sequence consists of a set of time-stamps, each of which records the arrival time of an event. Time-sequence data can generally be classified into two types. One is from experiments that monitor subjects in a continuous fashion; and thereby the exact time-stamps of all occurrences of the events are fully observable. These data are usually referred to as recurrent event data. On the other hand, we have the so-called panel count data, in which only the numbers of occurrences of the events between subsequent observation times. In real-world problems arising in areas such as social science, health care and crime prevention, time-sequence modeling is extremely useful since it can help us in predicting future events and understanding the reasons behind them. In this talk, I will discuss how to model multiple time-sequences with variational Bayesian inference when we have panel count data or recurrent event data.