Abstract:
Many of the breakthroughs of the last years in the applied sciences have been due to recent advances of machine learning, and in particular to deep learning. However, these methods have been shown to be unstable and susceptible to adversarial perturbations and deformations: a small error in the input may yield a large error in the output. In this talk, after a brief review of these phenomena, I will discuss its implications to the use of machine learning methods in inverse problems and imaging.