Temporal graphs, which model interactions over time, are essential for analyzing datasets in fields like Industry 4.0 and cybersecurity, but their sparsity and irregularity make change point (CP) detection challenging. Recent research suggests that using time-series and graph dictionaries to transform temporal graphs into new dynamical-structural domains can improve CP detection by highlighting key patterns. This internship focuses on optimizing dictionary selection to maximize detection accuracy, starting with established analytic dictionaries like Haar, Walsh, and Boolean-based ones for time-series data and exploring custom motif-based graph dictionaries to suit the sparse, binary nature of temporal graphs.
Temporal graphs, representing interactions over time, are crucial for analyzing datasets in areas like Industry 4.0, cybersecurity, and social networks. Temporal graphs often exhibit distinct activity regimes, making change point (CP) detection vital for tasks such as fault detection and prediction. However, the sparsity and irregularity of real-world temporal graphs make CP detection highly challenging, as current algorithms struggle to extract accurate patterns.
A recent set of works have proposed the use of time-series and graph dictionaries as a means to perform spectral transforms of temporal graphs, representing them in new dynamical-structural domains. Working in such transformed space shows potential for improving CP detection, making it easier to identify similarities and differences across different periods of a temporal graphs. Yet a key challenge must be addressed: determining the appropriate dictionaries so as to maximize detection accuracy.
The internship aims to tackle this question. While we are open to explore an end-to-end dictionary learning pipeline, our initial focus will be on evauating well-established analytic dictionaries, which allow theoretical bounds on detection rates. For time-series dictionaries we plan to conduct comparisons among Haar, Walsh, and Boolean-based dictionaries that are better adapted to the binary and sparse nature of temporal graphs. For graph dictionaries, we plan to build custom dictionaries with user-defined motifs.
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.
To apply, please send an e-mail to esteban.bautista@univ-littoral.fr, matthieu.puigt@univ-littoral.fr and laurent.brisson@imt-atlantique.fr while attaching the documents that can support your application: your resume; a cover letter; your transcripts from the last year of B.Sc to the last year of M.Sc. (if the latter is already available); two reference letters or the names and means of contact of two academic advisers.
Applications will be reviewed on a rolling basis until the position is filled.
(c) GdR IASIS - CNRS - 2024.