Speakers
Description
Machine Learning (ML) has seen limited adoption within large-scale networks (e.g. NRENs). Organisations are reluctant to share their data in fear of compromising end-user privacy, thus representative datasets to train accurate ML classifiers are usually not available. Moreover, complex black-box ML classifiers are not intrinsically explainable, hence network engineers are reluctant to deploy them. We present FedXAI4DNS that employs ML, Federated Learning (FL) and eXplainable AI (XAI) for collaborative and trustworthy detection of malignant DNS traffic produced by Domain Generation Algorithms (DGAs). FL enables collaborating organisations to jointly train privacy-aware classifiers without exchanging sensitive data, whereas XAI suggests methods for justifying configurations of complex black-box models. FedXAI4DNS aims at expediting ML adoption within collaborative environments (e.g. NRENs & GÉANT).
