Learning linearized assignment flows for image labeling
From MaRDI portal
Ordinary differential equations and systems on manifolds (34C40) Image analysis in multivariate analysis (62H35) Computer science (68-XX) Learning and adaptive systems in artificial intelligence (68T05) Computing methodologies for image processing (68U10) Evolutionary games (91A22) Information and communication theory, circuits (94-XX)
Abstract: We introduce a novel algorithm for estimating optimal parameters of linearized assignment flows for image labeling. An exact formula is derived for the parameter gradient of any loss function that is constrained by the linear system of ODEs determining the linearized assignment flow. We show how to efficiently evaluate this formula using a Krylov subspace and a low-rank approximation. This enables us to perform parameter learning by Riemannian gradient descent in the parameter space, without the need to backpropagate errors or to solve an adjoint equation. Experiments demonstrate that our method performs as good as highly-tuned machine learning software using automatic differentiation. Unlike methods employing automatic differentiation, our approach yields a low-dimensional representation of internal parameters and their dynamics which helps to understand how assignment flows and more generally neural networks work and perform.
Cites work
- Algorithm 919: A Krylov subspace algorithm for evaluating the -functions appearing in exponential integrators
- Analysis of Some Krylov Subspace Approximations to the Matrix Exponential Operator
- Approximation of functions of large matrices with Kronecker structure
- Assignment flows
- Assignment flows for data labeling on graphs: convergence and stability
- Computing low‐rank approximations of the Fréchet derivative of a matrix function using Krylov subspace methods
- Computing the action of the matrix exponential, with an application to exponential integrators
- Exponential integrators
- Exponential Rosenbrock-Type Methods
- Functions of Matrices
- Geometric numerical integration of the assignment flow
- scientific article; zbMATH DE number 3783337 (Why is no real title available?)
- scientific article; zbMATH DE number 194139 (Why is no real title available?)
- scientific article; zbMATH DE number 1953444 (Why is no real title available?)
- scientific article; zbMATH DE number 1795895 (Why is no real title available?)
- scientific article; zbMATH DE number 6982909 (Why is no real title available?)
- Image labeling by assignment
- Learning linear assignment flows for image labeling via exponential integration
- Nineteen Dubious Ways to Compute the Exponential of a Matrix, Twenty-Five Years Later
- Numerical methods for large eigenvalue problems
- On Krylov Subspace Approximations to the Matrix Exponential Operator
- Ordinary differential equations and dynamical systems
- The ubiquitous Kronecker product
Cited in
(5)- Learning adaptive regularization for image labeling using geometric assignment
- Unsupervised assignment flow: label learning on feature manifolds by spatially regularized geometric assignment
- A Nonlocal Graph-PDE and Higher-Order Geometric Integration for Image Labeling
- Riemannian patch assignment gradient flows
- A novel Krylov subspace method for approximating Fréchet derivatives of large-scale matrix functions
This page was built for publication: Learning linearized assignment flows for image labeling
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6156048)
