Point of contact
Virgile Prevosto, CEA List
Summary
Deep learning and neural networks have been extremely successful in many different domains. However, when it comes to trust an AI algorithm for critical tasks (e.g. autonomous driving), the question of how to specify what it should do, and how to prove that it will behave according to this specification remains largely open.
A few references
- Digicosme Spring School on Formal Methods and Machine Learning, ForMal, 2019.
- Julien Girard-Satabin et al. CAMUS: A Framework to Build Formal Specifications for Deep Perception Systems Using Simulators. In 24th European Conference on Artificial Intelligence (ECAI). 2020
- LeaRNNify project