Geometric numerical integration of the assignment flow

DOI10.1088/1361-6420/AB2772zbMATH Open1490.65139arXiv1810.06970OpenAlexW2990578246WikidataQ127739613 ScholiaQ127739613MaRDI QIDQ5000560FDOQ5000560

Stefania Petra, Christoph Schnörr, Fabrizio Savarino, Alexander Zeilmann

Publication date: 14 July 2021

Published in: Inverse Problems (Search for Journal in Brave)

Abstract: The assignment flow is a smooth dynamical system that evolves on an elementary statistical manifold and performs contextual data labeling on a graph. We derive and introduce the linear assignment flow that evolves nonlinearly on the manifold, but is governed by a linear ODE on the tangent space. Various numerical schemes adapted to the mathematical structure of these two models are designed and studied, for the geometric numerical integration of both flows: embedded Runge-Kutta-Munthe-Kaas schemes for the nonlinear flow, adaptive Runge-Kutta schemes and exponential integrators for the linear flow. All algorithms are parameter free, except for setting a tolerance value that specifies adaptive step size selection by monitoring the local integration error, or fixing the dimension of the Krylov subspace approximation. These algorithms provide a basis for applying the assignment flow to machine learning scenarios beyond supervised labeling, including unsupervised labeling and learning from controlled assignment flows.

Full work available at URL: https://arxiv.org/abs/1810.06970

Recommendations

• Geometric integrators for ODEs
• Geometric integration algorithms on homogeneous manifolds
• scientific article; zbMATH DE number 4213749
• scientific article; zbMATH DE number 1745051
• Geometric Numerical Integration
• Geometric Numerical Integration
• scientific article; zbMATH DE number 1343130
• Geometric integration: numerical solution of differential equations on manifolds
• scientific article; zbMATH DE number 1734659
• Optimal geometric flows via dual programs

zbMATH Keywords

geometric integrationKrylov subspace approximationassignment manifoldimage labelingadaptive step size selectionassignment flow

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05) Image analysis in multivariate analysis (62H35) Image processing (compression, reconstruction, etc.) in information and communication theory (94A08) Iterative numerical methods for linear systems (65F10) Multistep, Runge-Kutta and extrapolation methods for ordinary differential equations (65L06) Numerical methods for initial value problems involving ordinary differential equations (65L05)

Cites Work

• Expokit
• Algorithm 919
• Functions of Matrices
• Nineteen Dubious Ways to Compute the Exponential of a Matrix, Twenty-Five Years Later
• Solving Ordinary Differential Equations I
• Geometric Numerical Integration
• Ordinary differential equations and dynamical systems
• On some inequalities for the incomplete gamma function
• Title not available (Why is that?)
• High order Runge-Kutta methods on manifolds
• Exponential integrators
• Computing the Action of the Matrix Exponential, with an Application to Exponential Integrators
• Title not available (Why is that?)
• On Krylov Subspace Approximations to the Matrix Exponential Operator
• On a notion of linear replicator equations
• Analysis of Some Krylov Subspace Approximations to the Matrix Exponential Operator
• Image Labeling Based on Graphical Models Using Wasserstein Messages and Geometric Assignment
• Information Geometry
• Image labeling by assignment
• Stable architectures for deep neural networks
• A proposal on machine learning via dynamical systems
• Iterative multiplicative filters for data labeling

Cited In (12)

• Learning linear assignment flows for image labeling via exponential integration
• On the correspondence between replicator dynamics and assignment flows
• Assignment flows for data labeling on graphs: convergence and stability
• Learning linearized assignment flows for image labeling
• Learning adaptive regularization for image labeling using geometric assignment
• Assignment Flows
• A Nonlocal Graph-PDE and Higher-Order Geometric Integration for Image Labeling
• On Multiscale Quasi-Interpolation of Scattered Scalar- and Manifold-Valued Functions
• Assignment flow for order-constrained OCT segmentation
• Continuous-domain assignment flows
• Title not available (Why is that?)
• On the geometric mechanics of assignment flows for metric data labeling

Uses Software

• Digital Library of Mathematical Functions
• Expokit
• Algorithm 919
• phipm
• mftoolbox

This page was built for publication: Geometric numerical integration of the assignment flow

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5000560)