Generating corner cases for crashtesting deep networks

Jordan Platon (1) , Guillaume Avrin (2) , Adrien Chan-Hon-Tong (3)

1 OP - Observatoire de Paris - Site de Paris
2 LNE - Laboratoire National de Métrologie et d’Essais
3 DTIS, ONERA, Université Paris Saclay [Palaiseau]

Today, adversarial attack is almost the only kind of hard samples considered by the community to tickle deep networks. However, such additive perturbations are not realistic for many applications (e.g. remote sensing). This paper introduce a new kind of hard samples called corner cases where the entire data is generated (not just a perturbation). This allows to constraint more easily those data to be both hard but also realistic.

Domaines: Traitement des images [eess.IV]; Intelligence artificielle [cs.AI]

Jordan Platon, Guillaume Avrin, Adrien Chan-Hon-Tong. Generating corner cases for crashtesting deep networks. ECAI, Aug 2020, online, Spain. ⟨hal-01883078v3⟩