Explainable Anomaly Detection for Safeguarding and Enhancing Modern Data Industry

Description du projet

Système explicable de détection d’anomalies pour l’industrie des données

L’expansion rapide des technologies numériques et de l’internet des objets a entraîné une dépendance croissante à l’égard des technologies, des flux et du stockage des données. Cependant, à mesure que les volumes de données augmentent, les anomalies, les bogues et autres problèmes deviennent plus fréquents, ce qui nécessite des solutions avancées de détection et de prévention. Le projet ExplainableAD, financé par le CER, vise à relever les défis de la détection d’anomalies explicables (XAD) en développant un système XAD innovant qui renforce la sécurité des services numériques. Le projet intégrera la détection d’anomalies basée sur l’apprentissage profond avec des méthodes de découverte d’explications, fournissant aux utilisateurs des outils puissants pour identifier les activités frauduleuses et les dysfonctionnements de manière efficace et transparente.

Objectif

In the ExplainableAD project, we address the problem of eXplainable Anomaly Detection (XAD) on data streams with evolving, previously unknown, complex anomaly types. To bridge the gap between growing demands in the data industry and severely limited industry-grade solutions available today, we propose a breakthrough XAD system that offers unprecedented capabilities for safeguarding digital services. To this end, it seamlessly integrates deep learning-based anomaly detection (AD), which unlocks the power of detecting evolving, previously unknown, complex anomaly types, and explanation discovery (ED), which overcomes the non-interpretability of deep learning by returning human-readable, actionable insights. To establish a pathway from fundamental research to innovation, we will conduct 1) additional research activities on automated model selection, scalable execution, and data visualization; (2) prototyping, benchmarking analysis, and system demonstration; (3) large-scale use case studies with industry partners, leveraging ongoing collaborations and building new use cases across banking, finance, and healthcare domains; (4) market analysis and market entry strategy development to enable technology transfer.

Our XAD system will have a direct impact on end users across a variety of sectors, including fraud analysts in banking and financial services to combat ever-growing, complex fraud types, IT departments and cloud services to manage malfunction and performance challenges proactively, and biomedical researchers and staff to benefit from enhanced diagnostic accuracy and improved confidence. The PI will bring extensive experience from the ERC CoG grant, including a large number of publications and a highly skilled team, to this project. With a proven track record of delivering cutting-edge solutions to critical challenges faced by enterprises and society, she will leverage her strong industry connections to facilitate the transition from research to innovation.