Webinaire – Vendredi 13 janvier de 15h à 16h
Machine learning algorithms are increasingly used by critical systems such as autonomous vehicles or surgical robots due to their high capacity to learn patterns in complex data.
However, these algorithms, especially in computer vision, can output wrong answers even with a high confidence value associated with this answer.
In this talk, we will learn about SimOOD, an evolutionary approach for finding and reproducing common visual perturbations that may lead to failures in ML models at simulation time.
We will learn about the architecture, the first results, and how to use SimOOD along with the CARLA simulator, an open-source simulator for development, training, and validation of autonomous driving systems.
The presentation will be given by Raul Sena Ferreira.