Abstract

Compared with time series analysis, changepoint detection in point processes has received considerably less attention, despite its importance in a wide range of applications. Unlike conventional time series, point process data consist of events occurring at random times, with event timestamps often being the only available observations. This fundamental difference introduces distinct statistical and computational challenges for identifying changes in the underlying stochastic process. In this talk, I will discuss these challenges and introduce a statistical framework for detecting multiple changepoints in point processes, with particular emphasis on inhomogeneous Poisson processes. I will present the proposed methodology and its computationally efficient implementation for identifying changes in the underlying point process. Through applications to neural spike-train data recorded from ensembles of neurons, I will show how changepoint detection for point processes can reveal transitions in neural activity and provide insights into the temporal dynamics of neuronal populations.