Offre de stage : From Deep learning based Graph Anomaly Detection to Immersive Explanations in Augmented Reality

Duration: 6 months
Supervision: Jean-Philippe Farrugia (LIRIS, Lyon 1), Mohamed-Lamine Messai
(ERIC, Lyon 2) and Hamida SEBA (LIRIS, Lyon 1)
Location: LIRIS, campus de la Doua Villeurbanne or IUT Lyon 1, campus de Bourg en Bresse
Application should be sent to: Hamida.seba@univ-lyon1.fr (CV+Transcripts)
Keywords: Graph Anomaly Detection, Graph Neural Networks, Explainable AI, Graph-XAI, Augmented Reality, WebXR, Cybersecurity, Immersive Analytics.

Context

Cybersecurity systems generate large volumes of interconnected data: communications between machines, authentication events, processes, users, resources, and cyber-physical dependencies can naturally be represented as graphs. Graph Anomaly Detection (GAD) aims to identify unusual nodes, edges, subgraphs, or graphs. Deep learning approaches, and Graph Neural Networks (GNNs) in particular, have recently become an important paradigm for Graph Anomaly Detection because of their ability to jointly exploit graph structure and node or edge attributes.
A deep anomaly detector can typically associate an entity v with an anomaly score : fθ(G, v) → s(v).

However, knowing that s(v) is high is generally insufficient for a cybersecurity analyst. The analyst also needs to understand why the entity was classified as anomalous and which surrounding entities, attributes, interactions, or structural patterns contributed to the decision.

Recent research has started to investigate interpretable and explainable Graph Anomaly Detection. However, producing an algorithmic explanation does not guarantee that this explanation can be easily understood by a human analyst. At the same time, Augmented Reality offers new opportunities for presenting complex graph structures spatially and interactively.
This internship therefore investigates the connection between Deep Graph Anomaly Detection, Explainable AI, and Augmented Reality. The central research question is: How can an explanation produced by a deep graph anomaly detector be transformed into a perceptible, interactive, and useful spatial explanation for a cybersecurity analyst in Augmented Reality?
The project will reuse and extend the existing WebXR graph prototype, allowing the student to focus primarily on the scientific relationship between explainability and immersive representation rather than on the development of an AR platform from scratch.

Recent research has started to investigate interpretable and explainable Graph Anomaly Detection. However, producing an algorithmic explanation does not guarantee that this explanation can be easily understood by a human analyst.
At the same time, Augmented Reality offers new opportunities for presenting complex graph structures spatially and interactively.
This internship therefore investigates the connection between Deep Graph Anomaly Detection, Explainable AI, and Augmented Reality.
The central research question is:
How can an explanation produced by a deep graph anomaly detector
be transformed into a perceptible, interactive, and useful spatial
explanation for a cybersecurity analyst in Augmented Reality?
The project will reuse and extend the existing WebXR graph prototype, allowing the student to focus primarily on the scientific relationship between explainability and immersive representation rather than on the development of an AR platform from scratch.