Stability of the scattering transform for deformations with minimal regularity

DOI10.1016/J.MATPUR.2023.10.008MaRDI QIDQ6139821zbMATH OpenFDO

Authors Fabio Nicola, S. Ivan Trapasso

Publication date 19 December 2023

Published in Journal de Mathématiques Pures et Appliquées. Neuvième Série (Search for Journal in Brave)

Copyright license Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International

Full work available at URL https://arxiv.org/abs/2205.11142

zbMATH Keywords

deformationsmathematics of artificial intelligencemultiresolution analysisscattering transformstabilitywavelets

Mathematics Subject Classification ID

Function spaces arising in harmonic analysis (42B35) General harmonic expansions, frames (42C15) Nontrigonometric harmonic analysis involving wavelets and other special systems (42C40) Learning and adaptive systems in artificial intelligence (68T05) Artificial neural networks and deep learning (68T07) Signal theory (characterization, reconstruction, filtering, etc.) (94A12)

Abstract: Within the mathematical analysis of deep convolutional neural networks, the wavelet scattering transform introduced by St'ephane Mallat is a unique example of how the ideas of multiscale analysis can be combined with a cascade of modulus nonlinearities to build a nonexpansive, translation invariant signal representation with provable geometric stability properties, namely Lipschitz continuity to the action of small $C^2$ diffeomorphisms - a remarkable result for both theoretical and practical purposes, inherently depending on the choice of the filters and their arrangement into a hierarchical architecture. In this note, we further investigate the intimate relationship between the scattering structure and the regularity of the deformation in the H"older regularity scale $C^{a}lpha$, $alpha>0$. We are able to precisely identify the stability threshold, proving that stability is still achievable for deformations of class $C^{alpha}$, $alpha>1$, whereas instability phenomena can occur at lower regularity levels modelled by $C^{a}lpha$, $0lealpha<1$. While the behaviour at the threshold given by Lipschitz (or even $C^1$) regularity remains beyond reach, we are able to prove a stability bound in that case, up to $varepsilon$ losses.

Recommendations

• On generalizations of the nonwindowed scattering transform
• Group invariant scattering
• The Scattering Transform
• Study of a robust feature: the pointwise Lipschitz regularity
• Properties of the scattering transform on the real line

Cites work

• A Mathematical Theory of Deep Convolutional Neural Networks for Feature Extraction
• A wavelet tour of signal processing. The sparse way.
• Analysis of time-frequency scattering transforms
• Deep double descent: where bigger models and more data hurt*
• Group invariance, stability to deformations, and complexity of deep convolutional representations
• Group invariant scattering
• scientific article; zbMATH DE number 412139 (Why is no real title available?)
• scientific article; zbMATH DE number 447275 (Why is no real title available?)
• scientific article; zbMATH DE number 4175772 (Why is no real title available?)
• scientific article; zbMATH DE number 3324255 (Why is no real title available?)
• scientific article; zbMATH DE number 3345794 (Why is no real title available?)
• Interpolation non linéaire et régularité
• Interpolation theory
• Lipschitz properties for deep convolutional networks
• New tight frames of curvelets and optimal representations of objects with piecewise C² singularities
• On Lipschitz Bounds of General Convolutional Neural Networks
• Reconciling modern machine-learning practice and the classical bias-variance trade-off
• Shapes and diffeomorphisms
• Short Communication: Localized Adversarial Artifacts for Compressed Sensing MRI
• Sparse multidimensional representations using anisotropic dilation and shear operators
• Surprises in high-dimensional ridgeless least squares interpolation
• The curvelet representation of wave propagators is optimally sparse
• The Generalization Error of Random Features Regression: Precise Asymptotics and the Double Descent Curve
• Theory of Besov spaces
• Theory of function spaces
• Variational methods in imaging

Cited in

(5)

• On generalizations of the nonwindowed scattering transform
• Manifold filter-combine networks
• Generalized moduli of continuity under irregular or random deformations via multiscale analysis
• Energy propagation in scattering convolution networks can be arbitrarily slow
• Multiscale Hodge scattering networks for data analysis

This page was built for publication: Stability of the scattering transform for deformations with minimal regularity

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6139821)