This tutorial will discuss in detail attack detection through unsupervised anomaly detection. It will review the construction of an evaluation campaign through i) the identification of the attack models and datasets, ii) the selection and discussion of unsupervised algorithms, even when iii) adopting meta-learning, iv) the identification of target metrics, v) the execution of the algorithms and vi) their comparison.
Attendees will also be involved in an hands-on session where algorithms will be executed on public attack datasets thanks to RELOAD, a tool for the Rapid EvaLuation Of Anomaly Detection algorithms. The tool is primarily meant to be used by non-experts that start approaching binary classification using ML algorithms, and hides details which may be misleading for beginners, executing analyses through a simple UI.
SWT Leaders:
Tommaso Zoppi (Primary Contact)
University of Florence
tommaso.zoppi@unifi.it
Andrea Ceccarelli
University of Florence
andrea.ceccarelli@unifi.it
Andrea Bondavalli
University of Florence
bondavalli@unifi.it