A Note on the Regularity of Images Generated by Convolutional Neural Networks

DOI10.1137/22M1525995MaRDI QIDQ6136232zbMATH OpenOpenAlexFDO

Authors Andreas Habring, M. Holler

Publication date 29 August 2023

Published in SIAM Journal on Mathematics of Data Science (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2204.10588

zbMATH Keywords

machine learningfunctional analysisconvolutional neural networksmathematical imaging

Mathematics Subject Classification ID

Artificial neural networks and deep learning (68T07) Image processing (compression, reconstruction, etc.) in information and communication theory (94A08) Numerical solutions of ill-posed problems in abstract spaces; regularization (65J20)

Abstract: The regularity of images generated by 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 practical consequence, the results of this paper in particular provide analytical evidence that basic L2 regularization of network weights might lead to over-smoothed outputs.

Recommendations

• Regularization by architecture: a deep prior approach for inverse problems
• Big in Japan: regularizing networks for solving inverse problems
• Approximation properties of deep ReLU CNNs
• Lipschitz properties for deep convolutional networks
• A Generative Variational Model for Inverse Problems in Imaging

Cites work

• scientific article; zbMATH DE number 1865939 (Why is no real title available?)
• scientific article; zbMATH DE number 1448982 (Why is no real title available?)
• scientific article; zbMATH DE number 5681750 (Why is no real title available?)
• scientific article; zbMATH DE number 2208228 (Why is no real title available?)
• A Generative Variational Model for Inverse Problems in Imaging
• A convex variational model for learning convolutional image atoms from incomplete data
• A first course in Sobolev spaces
• An introduction to total variation for image analysis
• Are natural images of bounded variation?
• Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
• Deep Convolutional Neural Network for Inverse Problems in Imaging
• Deep synthesis network for regularizing inverse problems
• Functional analysis, Sobolev spaces and partial differential equations
• Linear functional analysis. An application-oriented introduction. Translated from the 6th German edition by Robert Nürnberg
• Mathematical image processing. Translated from the German
• Nonlinear total variation based noise removal algorithms
• Solving inverse problems using data-driven models

Cited in

(6)

• Image augmentation with conformal mappings for a convolutional neural network
• Continuous Generative Neural Networks: A Wavelet-Based Architecture in Function Spaces
• Neural-network-based regularization methods for inverse problems in imaging
• Inverse evolution layers: physics-informed regularizers for image segmentation
• Learning in the machine: to share or not to share?
• Multilevel diffusion: infinite dimensional score-based diffusion models for image generation

This page was built for publication: A Note on the Regularity of Images Generated by Convolutional Neural Networks

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6136232)