Temporal Logic for Learning and Detection of Anomalous behaviours

This repository contains the material relating to the seminar held for the Verification and Validation Techniques for AI & Cybersecurity course at the University of Udine.

Contents

The contribution was to review the work done by Kong, Jones and Belta in ¹. In this work, three algorithms are presented, based on iPSTL, a fragment of Parametric Signal Temporal Logic:

• the first is an Offline Anomaly Learning algorithm;
• the second is an Anomaly Detection algorithm;
• the third is an Online Anomaly Learning algorithm.

Finally, to test the proposed algorithms, two case studies are shown, one in the field of maritime surveillance and one for the control of a train braking system. The results obtained were excellent, in both scenarios, from all the algorithms.

Footnotes

1. [Kong] Zhaodan Kong, Austin Jones and Calin Belta. Temporal Logics for Learning and Detection of Anomalous Behavior. IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 62, NO. 3, MARCH 2017. ↩