Neural networks are increasingly being used for dealing with complex real-world applications. Despite their success, there are still important open issues such as their limited application in safety and security-critical contexts, wherein assurance about networks’ behavior must be provided. The development of reliable neural networks for safety-critical contexts is one of the topics investigated in the AIDOaRt Project, a 3 years long H2020-ECSEL European project focusing on Artificial Intelligence augmented automation supporting modeling, coding, testing, monitoring, and continuous development of Cyber-Physical Systems. In this paper, we present an interesting safety-critical use case – related to the automotive domain – from the AIDOaRt project. In addition, we outline the challenges we are facing in bridging the gap between the scalability of state-of-the-art verification methodologies and the complexity of the neural networks best suited for the task of interest.
Verification of Neural Networks: Challenges and Perspectives in the AIDOaRt Project / Eramo, R.; Fanni, T.; Guidotti, D.; Pandolfo, L.; Pulina, L.; Zedda, K.. - 3345:(2022). (Intervento presentato al convegno 10th Italian Workshop on Planning and Scheduling, IPS 2022, RCRA Incontri E Confronti, RiCeRcA 2022, and the Workshop on Strategies, Prediction, Interaction, and Reasoning in Italy, SPIRIT 2022 tenutosi a ita nel 2022).
Verification of Neural Networks: Challenges and Perspectives in the AIDOaRt Project
Fanni T.;Guidotti D.;Pandolfo L.;Pulina L.;
2022-01-01
Abstract
Neural networks are increasingly being used for dealing with complex real-world applications. Despite their success, there are still important open issues such as their limited application in safety and security-critical contexts, wherein assurance about networks’ behavior must be provided. The development of reliable neural networks for safety-critical contexts is one of the topics investigated in the AIDOaRt Project, a 3 years long H2020-ECSEL European project focusing on Artificial Intelligence augmented automation supporting modeling, coding, testing, monitoring, and continuous development of Cyber-Physical Systems. In this paper, we present an interesting safety-critical use case – related to the automotive domain – from the AIDOaRt project. In addition, we outline the challenges we are facing in bridging the gap between the scalability of state-of-the-art verification methodologies and the complexity of the neural networks best suited for the task of interest.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
