Abstract

This talk presents a novel framework aimed at improving failure detection in critical computing systems, where high reliability and safety are essential. It merges machine learning’s predictive power with the reliability of formal verification methods, using data streams from system operations, like telemetry, for early failure detection. The approach overcomes traditional verification challenges, such as specifying all properties of interest and fully modeling systems, by employing a lightweight runtime verification technique called monitoring, which bypasses the need for explicit model specifications. The integration of machine learning allows for the direct identification of failure patterns from system traces, represented as logical formulas, enabling effective real-time system verification and providing natural interpretability. The seminar will explore the framework’s development, from its conceptual basis to its prototype and potential extensions, while underscoring its implications for cutting-edge interdisciplinary research.

Bio 1: Andrea Brunello got his PhD in Computer Science in 2020 from the University of Udine (Italy). He is now an Assistant Professor at the Humanities Department of the same University. His main research interests are in data modelling, applied machine learning and on the integration between learning and formal methods.

Bio 2: Nicola Saccomanno is a Postdoc at the Department of Mathematics, Computer Science, and Physics of the University of Udine (Italy), where he got his PhD in 2023. His main research interests are in artificial intelligence, specifically symbolic and sub-symbolic integration, and applied machine learning, focusing on indoor positioning and healthcare domains.