A roadmap for curvature-based geometric data analysis and learning
Yasharth Yadav, Kelin Xia · 2025
Read A roadmap for curvature-based geometric data analysis and learning by Yasharth Yadav, Kelin Xia (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-22599.txt.
Introduction
How much a shape bends is called curvature. That bend can reveal hidden traits in data. Yasharth Yadav and Kelin Xia wrote a 2025 review on this theme. It covers the use of curvature in data work and in learning. The data types span graphs, point clouds, and cell complexes.
The review lays out the math base for each model. It also lays out how to compute with them. Uses span group finding in networks and deep learning on shapes. The pair view curvature from two angles of geometry. They set out a clear pipeline for such work. Work from the top groups is summed up at the end. Open paths for the field close the review.
Catalog record
| CREATOR | Yasharth Yadav, Kelin Xia |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 132 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-22599
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A roadmap for curvature-based geometric data analysis and learning Thanks: This work is supported by Singapore Ministry of Education Academic Research fund Tier 1 grant RG16/23 and Tier 2 grants MOE-T2EP20221-0003, MOE-T2EP50223-0014. Yasharth Yadav, Kelin Xia Abstract Geometric data analysis and learning has emerged as a distinct and rapidly developing research area, increasingly recognized for its effectiveness across diverse applications. At the heart of this field lies curvature, a powerful and interpretable concept that captures intrinsic geometric structure and underpins numerous tasks, from community detection to geometric deep learning. A wide range of discrete curvature models have been proposed for various data representations, including graphs, simplicial complexes, cubical complexes, and point clouds sampled from manifolds. These models not only provide efficient characterizations of data geometry but also constitute essential components in geometric learning frameworks. In this paper, we present the first comprehensive review of existing discrete curvature models, covering their mathematical foundations, computational formulations, and practical applications in data analysis and learning. In particular, we discuss discrete curvature from both Riemannian and metric geometry perspectives and propose a systematic pipeline for curvature-driven data analysis. We further examine the corresponding computational algorithms across different data representations, offering detailed comparisons and insights. Finally, we review state-of-the-art applications of curvature in both supervised and unsupervised learning. This survey provides a conceptual and practical roadmap for researchers to gain a better understanding of discrete curvature as a fundamental tool for geometric understanding and learning. keywords Discrete Curvature, Ricci Curvature, Sectional Curvature, Discrete Ricci Flow, Geometric Data Analysis, Geometric Deep Learning † † runningheads: Curvature-based geometric data analysis & learning / Y. Yadav, & K. Xia MSC 05C10, 51F99, 53C44, 68T30, 97N70 1 Introduction Recent advances in machine learning have led to powerful methods for tackling a broad range of problems spanning domains such as computer vision [ 160 ] , natural language processing [ 137 , 123 ] , and speech recognition [ 128 ] . Most of these methods are applicable on data with an underlying Euclidean structure, where data points are represented as vectors within a high-dimensional Euclidean space. However, modern applications increasingly rely on data that exhibits a non-Euclidean structure. Prominent examples include social networks in computational social science [ 100 ] , sensor networks in wireless systems [ 176 ] , functional connectomes derived from brain imaging [ 144 ] , and gene regulatory networks in systems biology [ 40 ] . Geometric deep learning aims to construct learning frameworks on non-Euclidean datasets [ 24 , 29 , 183 ] . At its core, geometric deep learning centers on incorporating the structure of data directly into model architectures. One key development in geometric deep learning is the design of graph convolutional networks (GCNs) [ 94 ] , which generalize convolutional neural networks (CNNs) to graph-structured data. CNNs have achieved strong success in image processing [ 104 ] due to their ability to leverage translation invariance as an inductive bias, enabling them to detect patterns in images regardless of spatial position. However, in non-Euclidean domains such as graphs, there is no natural notion of translation. Therefore, the standard convolution kernel fails to capture the same structural information. GCNs address this limitation by aggregating information from the local neighborhood of a node, effectively capturing both node features and local connectivity to learn useful representations. This formulation has led GCNs to achieve state-of-the-art results across a wide range of graph learning tasks [ 198 ] . Other approaches to generalize CNNs to graphs include the use of recurrent neural networks combined with random walks [ 69 , 169 ] , as well as convolutions defined in the spectral domain [ 25 ] . A central theme across these models is their dependence on the structure and connectivity of data. As such models gain prominence, it becomes essential to develop tools that can formally characterize this connectivity. Discrete notions of curvature provide scalable and interpretable methods for quantifying the geometry of diverse data representations, including graphs, simplicial complexes, cubical complexes, and point clouds sampled on manifolds. In Riemannian geometry, curvature is a local invariant of a manifold that measures its deviation from being Euclidean [ 180 , 127 , 87 ] . The notion of curvature has also been extended beyond smooth manifolds to more general metric spaces, often with a discrete structure [ 167 , 119 , 21 ] . A key qualitative feature shared by many of these discrete curvature models is their ability to distinguish bottleneck regions from highly clustered regions. Nevertheless, each definition is fundamentally grounded in a distinct mathematical principle, reflecting a specific property of classical Riemannian curvature. Among the most prominent definitions are Forman’s discretization based on a combinatorial analog of the classical Bochner Weitzenböck formula in differential geometry [ 60 ] , Ollivier’s coarse Ricci curvature for metric spaces equipped with a measure or random walk [ 133 , 134 ] , and Bakry-Émery’s generalized lower bound of Ricci curvature [ 6 ] . Other definitions include Menger [ 122 ] and Haantjes [ 73 ] curvatures which capture geometric properties of triangles and paths in metric spaces respectively, curvature based on effective resistance inspired by electrical circuit theory [ 43 ] , and a notion of sectional curvature that provides a global perspective on geometry [ 83 , 84 ] . While many of these definitions have been studied theoretically [ 9 , 10 , 88 ] , often in isolation, a few have also been evaluated empirically, primarily in the context of network data [ 164 , 163 , 148 , 124 ] . These notions of discrete curvature have also found several applications in machine learning. One of the earliest and most influential applications involves unsupervised node clustering or community detection in graphs, where two main strategies have been proposed, namely incremental removal of negatively curved edges [ 158 , 57 ] and updating edge weights using an associated discrete Ricci flow [ 132 , 98 ] . More recently, discrete curvatures have gained prominence in deep learning architectures, particularly in graph neural networks (GNNs) [ 156 , 41 ] . Notably, discrete curvatures have been used to characterize two central challenges in training graph neural networks, namely oversmoothing and oversquashing [ 109 , 3 ] . Several curvature-informed graph rewiring methods have also been proposed as preprocessing steps that modify the graph structure to improve learning performance [ 175 , 131 , 59 ] . The growing number of discrete curvature models and their expanding use in machine learning have resulted in a diverse but disconnected body of work that can be difficult to navigate for a newcomer. This paper provides a structured overview of key curvature models, their computation across different data representations, and their state-of-the-art applications in machine learning. Our goal is to support researchers and practitioners in understanding how curvature characterizes data geometry and how it can inform the design of more effective learning algorithms. The current paper is structured as follows. Section 2 presents widely used generalizations of curvature to general metric spaces, with particular emphasis on extensions of sectional, Ricci, and scalar curvature, and their theoretical motivations in Riemannian and metric geometry. Building on this foundation, Section 3