Change Point Detection in Temporal Graphs

Le LISIC développe des activités de recherche en sciences et technologies de l’information et plus spécifiquement dans les domaines de l’informatique et du traitement du signal.

Le laboratoire développe des recherches dans différents domaines de l'informatique et du traitement du signal. Il est composé de 4 équipes de recherche : IC (Ingénierie des connaissances), OSMOSE (Optimisation Simulation MOdeliSation Evolutionnaire), IMAP (Images et Apprentissage) et SPECIFI (Systèmes de Perception et Fusion d’Informations). Du point de vue de l'image, l'équipe IMAP développe des recherches en synthèse et analyse d'images, avec des applications en biologie, histoire et histoire de l'art.

Context:
Temporal graphs have emerged as a powerful framework for modeling dynamic interactions in various domains, from social networks to cybersecurity and Industry 4.0. These graphs capture temporal relationships between entities through triplets (u, v, t), indicating that entities u and v interacted at time t. What makes these structures particularly interesting is their ability to exhibit distinct activity regimes - for instance, social network interactions may follow different patterns during weekdays versus weekends, or banking transactions may show varying behaviors during different times of day. However, the inherent challenges of temporal graphs, including their high sparsity and irregular nature, make it exceptionally difficult to reliably detect these regime changes, limiting the effectiveness of current detection algorithms despite recent advances in the field.

Objectives:
This 6-month research internship aims to advance the state-of-the-art in change point detection for temporal graphs through a novel dictionary-based approach. The primary focus will be on developing and implementing two complementary strategies: First, exploring established analytic dictionaries (including Haar, Walsh, and boolean-based) for time-series analysis, with the goal of establishing theoretical bounds on detection rates. Second, creating custom graph dictionaries with user-defined motifs, employing a data-driven methodology to identify and incorporate the most relevant structural patterns. The project will emphasize both theoretical foundations and practical implementation, with particular attention to maximizing detection accuracy while minimizing false positives. The successful candidate will have the opportunity to work at the intersection of graph theory, signal processing, and machine learning, contributing to a emerging field with significant real-world applications.

Please find full offer description here: 
https://estbautista.com/wp-content/uploads/2024/11/Data2Laws___M2_Internship.pdf

This internship is directed at students with various backgrounds (computer science, data science, signal processing, complex systems) but with a strong interest in data science and graphs. Interest in the theoretical aspects of machine learning and in Python development will be a plus.