Anomaly and Change Detection in Sequences of Graphs

You are cordially invited to attend the PhD Dissertation Defence of Daniele Zambon on Thursday 16 December 2021 at 10:15 in room D0.02 or online on MS Teams.

Abstract:
We are experiencing a considerable growth in the amount of data gathered from sensor and social networks thanks to advances in mobile and distributed devices, as well as the spread of social platforms. Extracting information from such large datasets is crucial both to academia and industry. As sensor data streams not rarely share functional dependencies, graphs emerge as rich structures modeling both information at the sensor/entity level and the complex relations existing among entities. In turn, this representation has enabled inference from streams of graphs, for example, through graph neural networks and geometric deep learning. However, learning from real-world data often involves dealing with systems that change their operating conditions over time due to, e.g., sensor aging, new market behaviors, or change in the users' preferences. In this doctoral thesis, we address the problem of identifying changes in stationarity emerging in sequences of graphs caused by unknown phenomena occurring to the underlying data-generating process. This outcome permits us to also address the anomaly detection problem in sequences of graphs that indeed represents a valuable follow-up of the research. To cover the broadest spectrum of applications, we consider a general family of attributed graphs with non-identified vertices. The major contribution of the thesis includes a methodology for processing sequences of graphs in order to detect unexpected behaviors in the driving process (non-stationarities and/or anomalies). This passes through the design of novel rich graph-level embeddings and graph-based change detection methods supported by a novel and solid theoretical framework.

Dissertation Committee:
- Prof. Cesare Alippi, Università della Svizzera italiana, Switzerland (Research Advisor)
- Prof. Lorenzo Livi, University of Manitoba, Canada (Research co-Advisor)
- Prof. Michael Bronstein, Università della Svizzera italiana, Switzerland (Internal Member)
- Prof. Illia Horenko, Università della Svizzera italiana, Switzerland (Internal Member)
- Prof. Albert Bifet, University of Waikato, New Zealand and LTCI, Télécom Paris, France (External Member)
- Prof. Marco Gori, University of Siena, Italy and Université Côte d’Azur, France (External Member)
- Prof. Danilo Mandic, Imperial College London, UK (External Member)