About This Workshop
As machine learning-based systems are being deployed in safety-critical applications such as autonomous driving, medical imaging, or cyber-security systems, characterizing their behavior not only in the average but also worst case becomes essential. However, most existing research treats machine learning models such as deep neural networks as black boxes and uses simple empirical metrics such as their mean accuracy to quantify their performance. However, accuracy alone is not sufficient to assure that models conform to even basic safety or robustness specifications. To fill this gap, formal verification algorithms for machine learning aim to formally prove or disprove desired properties of machine learning models, including safety, fault tolerance, fairness, robustness, and correctness.
