Séance du 22 novembre 2021

Séance organisée par Madalina Olteanu et Fabrice Rossi. Séance consacrée à la détection de ruptures.

Lieu : IHP, amphi Darboux

14.00 : Rebecca Killick (Maths and Stats Department, Lancaster University)

Titre : A Circular Approach for Micro-Macro Changepoint Detection

Résumé : When considering data collected at higher resolutions, e.g. daily or sub-daily, we have to model the finer scale periodicities and/or changes that become part of the 'noise' to the signal we wish to understand. However, it can be hard to disentangle the large scale changes amongst the finer scale changes. In (sub-)daily climate measurements there are often periodic patterns that are present but we are not interested. In daily activity levels people often have a repeating pattern of activity but we are interested in more longer-term changes in behaviour. Failure to model the finer behaviour can lead to incorrect inference on the larger scale patterns. We propose a hierarchical circular changepoint approach that separately models the fine-scale changes from the longer-term changes. This can provide a more parsimonious representation of the data than traditional linear approaches. We demonstrate the approach on daily North Atlantic Oscillation data and passive monitoring of activity levels of elderly people in their homes.

15.00 : Laetitia Gauvin (Data Science Laboratory, ISI Foundation, Turin)

Titre : Representations for temporal networks: tensor based methods and embedding techniques

Résumé : There has been a recent surge in the representations of networks, especially in the development of methods to create embedding for networks that preserve important properties of the original structure, while representing it in a lower dimensional space. While these kind of methods were proved to be useful for anomaly detection or prediction of links in networks, they in general cannot be used directly for temporal networks. In this talk I will present several techniques to represent temporal networks and I will show how these representations can be used to learn relevant temporal and topological structures.

16.00 : Olga Klopp (CREST et ESSEC Business School)

Titre : Change-Point Detection in Dynamic Networks

Résumé : Structural changes occur in dynamic networks quite frequently and its detection is an important question in many applications. In this talk we consider the problem of change-point detection at a temporal sequence of partially observed networks. The goal is to test whether there is a change in the network parameters. Our approach is based on the Matrix CUSUM test statistic and allows growing size of networks. We propose a new test and show that it is minimax optimal and robust to missing links.