Guardado en:
Detalles Bibliográficos
Autores principales: Akande, Oluwatosin, Dondl, Patrick, Gupta, Kanan, Onwunta, Akwum, Wojtowytsch, Stephan
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:https://arxiv.org/abs/2501.00389
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909659481243648
author Akande, Oluwatosin
Dondl, Patrick
Gupta, Kanan
Onwunta, Akwum
Wojtowytsch, Stephan
author_facet Akande, Oluwatosin
Dondl, Patrick
Gupta, Kanan
Onwunta, Akwum
Wojtowytsch, Stephan
contents We study the momentum-based minimization of a diffuse perimeter functional on Euclidean spaces and on graphs with applications to semi-supervised classification tasks in machine learning. While the gradient flow in the task at hand is a parabolic partial differential equation, the momentum method corresponds to a damped hyperbolic PDE, leading to qualitatively and quantitatively different trajectories. Using a convex-concave splitting-based FISTA-type time discretization, we demonstrate empirically that momentum can lead to faster convergence if the time step size is large but not too large. With large time steps, the PDE analysis offers only limited insight into the geometric behavior of solutions and typical hyperbolic phenomena like loss of regularity are not be observed in sample simulations. We obtain the singular limit of the evolution equations as the length parameter of the phase fields tends to zero by formal expansions and numerically confirm its validity for circles in two dimensions. Our analysis is complemented by numerical experiments for planar curves, surfaces in three-dimensional space, and semi-supervised learning tasks on graphs.
format Preprint
id arxiv_https___arxiv_org_abs_2501_00389
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Momentum-based minimization of the Ginzburg-Landau functional on Euclidean spaces and graphs
Akande, Oluwatosin
Dondl, Patrick
Gupta, Kanan
Onwunta, Akwum
Wojtowytsch, Stephan
Analysis of PDEs
Numerical Analysis
Machine Learning
49Q05, 53E10, 35R02, 53Z50
We study the momentum-based minimization of a diffuse perimeter functional on Euclidean spaces and on graphs with applications to semi-supervised classification tasks in machine learning. While the gradient flow in the task at hand is a parabolic partial differential equation, the momentum method corresponds to a damped hyperbolic PDE, leading to qualitatively and quantitatively different trajectories. Using a convex-concave splitting-based FISTA-type time discretization, we demonstrate empirically that momentum can lead to faster convergence if the time step size is large but not too large. With large time steps, the PDE analysis offers only limited insight into the geometric behavior of solutions and typical hyperbolic phenomena like loss of regularity are not be observed in sample simulations. We obtain the singular limit of the evolution equations as the length parameter of the phase fields tends to zero by formal expansions and numerically confirm its validity for circles in two dimensions. Our analysis is complemented by numerical experiments for planar curves, surfaces in three-dimensional space, and semi-supervised learning tasks on graphs.
title Momentum-based minimization of the Ginzburg-Landau functional on Euclidean spaces and graphs
topic Analysis of PDEs
Numerical Analysis
Machine Learning
49Q05, 53E10, 35R02, 53Z50
url https://arxiv.org/abs/2501.00389