Guardat en:
Dades bibliogràfiques
Autors principals: Chen, Ziheng, Song, Yue, Wu, Xiao-Jun, Sebe, Nicu
Format: Preprint
Publicat: 2024
Matèries:
Accés en línia:https://arxiv.org/abs/2407.02607
Etiquetes: Afegir etiqueta
Sense etiquetes, Sigues el primer a etiquetar aquest registre!
_version_ 1866910014749278208
author Chen, Ziheng
Song, Yue
Wu, Xiao-Jun
Sebe, Nicu
author_facet Chen, Ziheng
Song, Yue
Wu, Xiao-Jun
Sebe, Nicu
contents Recent advances in Symmetric Positive Definite (SPD) matrix learning show that Riemannian metrics are fundamental to effective SPD neural networks. Motivated by this, we revisit the geometry of the Cholesky factors and uncover a simple product structure that enables convenient metric design. Building on this insight, we propose two fast and stable SPD metrics, Power--Cholesky Metric (PCM) and Bures--Wasserstein--Cholesky Metric (BWCM), derived via Cholesky decomposition. Compared with existing SPD metrics, the proposed metrics provide closed-form operators, computational efficiency, and improved numerical stability. We further apply our metrics to construct Riemannian Multinomial Logistic Regression (MLR) classifiers and residual blocks for SPD neural networks. Experiments on SPD deep learning, numerical stability analyses, and tensor interpolation demonstrate the effectiveness, efficiency, and robustness of our metrics. The code is available at https://github.com/GitZH-Chen/PCM_BWCM.
format Preprint
id arxiv_https___arxiv_org_abs_2407_02607
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Fast and Stable Riemannian Metrics on SPD Manifolds via Cholesky Product Geometry
Chen, Ziheng
Song, Yue
Wu, Xiao-Jun
Sebe, Nicu
Differential Geometry
Machine Learning
Metric Geometry
47A64, 26E60, 53C22, 15B48, 58D17, 53C20, 58B20
Recent advances in Symmetric Positive Definite (SPD) matrix learning show that Riemannian metrics are fundamental to effective SPD neural networks. Motivated by this, we revisit the geometry of the Cholesky factors and uncover a simple product structure that enables convenient metric design. Building on this insight, we propose two fast and stable SPD metrics, Power--Cholesky Metric (PCM) and Bures--Wasserstein--Cholesky Metric (BWCM), derived via Cholesky decomposition. Compared with existing SPD metrics, the proposed metrics provide closed-form operators, computational efficiency, and improved numerical stability. We further apply our metrics to construct Riemannian Multinomial Logistic Regression (MLR) classifiers and residual blocks for SPD neural networks. Experiments on SPD deep learning, numerical stability analyses, and tensor interpolation demonstrate the effectiveness, efficiency, and robustness of our metrics. The code is available at https://github.com/GitZH-Chen/PCM_BWCM.
title Fast and Stable Riemannian Metrics on SPD Manifolds via Cholesky Product Geometry
topic Differential Geometry
Machine Learning
Metric Geometry
47A64, 26E60, 53C22, 15B48, 58D17, 53C20, 58B20
url https://arxiv.org/abs/2407.02607