Saved in:
Bibliographic Details
Main Authors: Wang, Yanke, Flouris, Kyriakos
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.25206
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912615317372928
author Wang, Yanke
Flouris, Kyriakos
author_facet Wang, Yanke
Flouris, Kyriakos
contents This work explores optimization methods on hyperbolic manifolds. Building on Riemannian optimization principles, we extend the Hyperbolic Stochastic Gradient Descent (a specialization of Riemannian SGD) to a Hyperbolic Adam optimizer. While these methods are particularly relevant for learning on the Poincaré ball, they may also provide benefits in Euclidean and other non-Euclidean settings, as the chosen optimization encourages the learning of Poincaré embeddings. This representation, in turn, accelerates convergence in the early stages of training, when parameters are far from the optimum. As a case study, we train diffusion models using the hyperbolic optimization methods with hyperbolic time-discretization of the Langevin dynamics, and show that they achieve faster convergence on certain datasets without sacrificing generative quality.
format Preprint
id arxiv_https___arxiv_org_abs_2509_25206
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Hyperbolic Optimization
Wang, Yanke
Flouris, Kyriakos
Machine Learning
Computer Vision and Pattern Recognition
This work explores optimization methods on hyperbolic manifolds. Building on Riemannian optimization principles, we extend the Hyperbolic Stochastic Gradient Descent (a specialization of Riemannian SGD) to a Hyperbolic Adam optimizer. While these methods are particularly relevant for learning on the Poincaré ball, they may also provide benefits in Euclidean and other non-Euclidean settings, as the chosen optimization encourages the learning of Poincaré embeddings. This representation, in turn, accelerates convergence in the early stages of training, when parameters are far from the optimum. As a case study, we train diffusion models using the hyperbolic optimization methods with hyperbolic time-discretization of the Langevin dynamics, and show that they achieve faster convergence on certain datasets without sacrificing generative quality.
title Hyperbolic Optimization
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.25206