Towards Machine Learning Enhanced LTL Monitoring

In this work, we outline an extension of a recently proposed framework for failure detection that additionally supports the detection of anomalies and drops of performance of a given system. The extended framework is based on a tight integration of monitoring with unsupervised learning techniques, that are used to generate formulas able to capture possible deviations from the normal behaviour of the system or early signs of degradation phenomena. Other improvements to the framework are proposed like, for instance, the use of canonical forms for the safety and cosafety (monitorable) fragments of temporal logics and the support for assumption-based runtime verification.