Change-Point Detection in Dynamic Networks (deux exposés)

par Farida Enikeeva

mercredi 26 nov. 2025, 10:30 → 12:00 Europe/Paris

125 (Bat Braconnier, Doua)

Format 2*45 minutes

 

Change-Point Detection in Dynamic Networks

 

A dynamic network is a sequence of random graphs observed over time. Most real-life dynamic networks experience structural changes over time due to various factors such as evolving relationships, the addition or removal of nodes and edges, and external influences.  The detection of such changes is an important question in many practical situations, such as fraud detection or cybersecurity. The specific moments when these changes happen are called change points. In this talk, I will focus on statistical tests designed to identify changes in sequences of sparse high-dimensional graphs.

 

According to the minimax theory of statistical testing, the test performance is measured by the minimax separation rate. For the inhomogeneous random graph model, we establish the upper and lower bounds for the minimax separation rate and show the minimax rate optimality of our test, which is based on the Matrix CUSUM statistic.  We extend our results to the model of graphons and derive a nearly-minimax optimal test for K-step graphons. The results will be illustrated through a real-data example.