Deep neural networks on diffeomorphism groups for optimal shape reparametrization

DOI10.1007/S10543-023-00989-5MaRDI QIDQ6093564zbMATH OpenWikidataFDO

Authors Elena Celledoni, Helge Glöckner, Jørgen N. Riseth, Alexander Schmeding

Publication date 9 October 2023

Published in BIT (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2207.11141

zbMATH Keywords

deep learningshape analysisdiffeomorphism groupoptimal reparametrization

Mathematics Subject Classification ID

Numerical optimization and variational techniques (65K10) Groups of diffeomorphisms and homeomorphisms as manifolds (58D05) Manifolds of mappings (46T10)

Abstract: In shape analysis, one of the fundamental problems is to align curves or surfaces before computing a (geodesic) distance between these shapes. To find the optimal reparametrization realizing this alignment is a computationally demanding task which leads to an optimization problem on the diffeomorphism group. In this paper, we construct approximations of orientation-preserving diffeomorphisms by composition of elementary diffeomorphisms to solve the approximation problem. We propose a practical algorithm implemented in PyTorch which is applicable both to unparametrized curves and surfaces. We derive universal approximation results and obtain bounds for the Lipschitz constant of the obtained compositions of diffeomorphisms.

Recommendations

• Riemannian geometry on shapes and diffeomorphisms
• Riemannian optimization for registration of curves in elastic shape analysis
• Deep learning approximation of diffeomorphisms via linear-control systems
• Diffeomorphic learning
• A sub-Riemannian modular framework for diffeomorphism-based analysis of shape ensembles

Cites work

• scientific article; zbMATH DE number 3955268 (Why is no real title available?)
• scientific article; zbMATH DE number 3674997 (Why is no real title available?)
• scientific article; zbMATH DE number 5060482 (Why is no real title available?)
• scientific article; zbMATH DE number 7685246 (Why is no real title available?)
• A proposal on machine learning via dynamical systems
• Approximation by superpositions of a sigmoidal function
• Closed surfaces with different shapes that are indistinguishable by the SRNF.
• Constructing reparameterization invariant metrics on spaces of plane curves
• Control on the manifolds of mappings with a view to the deep learning
• Controllability on the group of diffeomorphisms
• Cubic convolution interpolation for digital image processing
• Deep learning as optimal control problems: models and numerical methods
• Density estimation by dual ascent of the log-likelihood
• Diffeomorphism groups of compact convex sets
• Infinite dimensional Lie transformation groups
• Regularity of Lie groups
• Shape analysis of surfaces using general elastic metrics
• Shape analysis on Lie groups and homogeneous spaces
• Shape analysis on Lie groups with applications in computer animation
• Stable architectures for deep neural networks
• Structure-preserving deep learning
• The blossoming of Schröder's fourth problem
• The inverse function theorem of Nash and Moser
• Towards a Lie theory of locally convex groups
• When unit groups of continuous inverse algebras are regular Lie groups

Cited in

(2)

• Designing stable neural networks using convex analysis and ODEs
• Constrained dynamics, stochastic numerical methods and the modeling of complex systems. Abstracts from the workshop held May 26--31, 2024

This page was built for publication: Deep neural networks on diffeomorphism groups for optimal shape reparametrization

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6093564)