A Note on the Regularity of Images Generated by Convolutional Neural Networks
Andreas Habring, Martin Höller · SIAM Journal on Mathematics of Data Science · 2023
Abstract. The regularity of images generated by a class of convolutional neural networks, such as the U-net, generative networks, or the deep image prior, is analyzed. In a resolution-independent, infinite dimensional setting, it is shown that such images, represented as functions, are always continuous and, in some circumstances, even continuously differentiable, contradicting the widely accepted modeling of sharp edges in images via jump discontinuities. While such statements require an infinite dimensional setting, the connection to (discretized) neural networks used in practice is made by considering the limit as the resolution approaches infinity. As a practical consequence, the results of this paper in particular provide analytical evidence that basic [Formula: see text] regularization of network weights (also known as weight decay) might lead to oversmoothed outputs.