Title: On Stability of Deep Learning in Image Reconstruction
Abstract: Deep learning has been proposed as a promising tool to increase performance for image reconstruction in medical imaging. In this talk we will introduce the image reconstruction problem, and show, with a lot of examples, how trained neural networks are completely unstable for image reconstruction. We will then present a theoretical result showing how deep learning will typically encourage this type of behavior for image reconstruction and inverse problems in general.
Download the flyer here