Giulio Rossolini, a researcher at the TeCIP Institute in the Cybersecurity area, has been awarded the 2025 Premio Giovani Ricercatrici e Ricercatori (Gruppo 2003), promoted by Scienza in Rete, for his work on evaluating the vulnerability of autonomous vehicle vision systems. The award particularly recognizes his contribution in proposing innovative solutions aimed at improving the resilience of these systems against cyberattacks.
The research, presented in the paper “On the Real-World Adversarial Robustness of Real-Time Semantic Segmentation Models for Autonomous Driving”, is part of the SERICS project and focuses on the security of machine learning systems for autonomous driving. The work introduces a new optimization method for generating physically realizable adversarial attacks and also proposes a defense strategy with probabilistic guarantees that balances robustness and computational cost. The study highlights a critical vulnerability of neural networks in real-world scenarios and shows how effective countermeasures can be designed, opening new lines of research in machine learning security.
