Abstract:In many data analysis tasks, one is often confronted with very high dimensional data. The manifold assumption, which states that the data is sampled from a submanifold embedded in much higher dimensional Euclidean space, has been widely adopted by many researchers. In the last 15 years, a large number of manifold learning algorithms have been proposed. Many of them rely on the evaluation of the geometrical and topological of the data manifold. In this paper, we present a review of these methods on a novel geometric perspective. We categorize these methods by three main groups: Laplacian-based, Hessian-based, and parallel field-based methods. We show the connection and difference between these three groups on their continuous and discrete counterparts. The discussion is focused on the problem of dimensionality reduction and semi-supervised learning.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the estimation and extraction of low - dimensional manifold structures in high - dimensional data. Specifically, the paper focuses on how to use manifold learning methods to reduce data dimensions while preserving the intrinsic topological and geometric properties of the data during this process. These methods are particularly important for dimensionality reduction and semi - supervised learning. ### Core Problems of the Paper 1. **Low - Dimensional Representation of High - Dimensional Data**: - There is often a low - dimensional intrinsic representation in high - dimensional data, that is, the data is actually sampled from a low - dimensional manifold embedded in a high - dimensional Euclidean space. Therefore, how to extract the structure of this low - dimensional manifold from high - dimensional data is a key issue. 2. **Classification and Comparison of Manifold Learning Methods**: - The paper classifies the existing manifold learning methods, mainly into Laplacian - based, Hessian - based, and parallel field - based methods. By comparing the continuous and discrete forms of these methods, it explores the relationships and differences between them. 3. **Applications in Dimensionality Reduction and Semi - Supervised Learning**: - Manifold learning is not only used for dimensionality reduction but also plays an important role in semi - supervised learning. The paper discusses how to use the manifold hypothesis and differential operators to construct regularization terms, thereby improving the performance of the model in semi - supervised learning. ### Key Concepts and Methods - **Laplacian**: It is used to measure the smoothness of a function and is often used for smooth regularization in dimensionality reduction and semi - supervised learning. - **Hessian**: It is used to measure how a function changes the metric of a manifold and is often used to preserve the linear or distance properties of data. - **Parallel Field**: By studying the parallelism of vector fields, it preserves the second - order smoothness of data and is suitable for tasks such as multi - task learning and manifold alignment. ### Main Contributions - **Review from a Geometric Perspective**: The paper systematically reviews manifold learning methods from a geometric perspective, providing new insights into these methods. - **Combination of Theory and Application**: It not only discusses the theoretical basis of the methods but also explores their applications in practical problems, such as dimensionality reduction and semi - supervised learning. - **Future Research Directions**: It points out the problems existing in current research and future research directions, such as the convergence of differential operator approximation and the theoretical explanation of vector field regularization. Through these discussions, the paper aims to provide a comprehensive perspective for researchers in the field of manifold learning and promote the development of this field.

A geometric viewpoint of manifold learning

Deep Manifold Computing and Visualization

Editorial of the Special Issue on Manifold Learning.

Deep Manifold Computing and Visualization Using Elastic Locally Isometric Smoothness

Deep Manifold Transformation for Nonlinear Dimensionality Reduction

Geodesic Based Semi-supervised Multi-manifold Feature Extraction

A Regularized Approach for Geodesic-Based Semisupervised Multimanifold Learning

Investigating Manifold Learning Algorithms Based on Magnification Factors and Principal Spread Directions

Riemannian manifold learning.

The Mathematical Foundations of Manifold Learning

Tangent Space Learning and Generalization

Manifold Learning for Visualizing and Analyzing High-Dimensional Data

Active learning on manifolds

Extendable and invertible manifold learning with geometry regularized autoencoders

Manifold learning: what, how, and why

Image Retrieval Algorithms Based on Manifold Learning

Geodesic Gabriel Graph Based Supervised Nonlinear Manifold Learning

A Comprehensive Survey on Geometric Deep Learning

Only Bayes should learn a manifold (on the estimation of differential geometric structure from data)

Learning High-Dimensional Correspondence Via Manifold Learning and Local Approximation

Exploring the Manifold of Neural Networks Using Diffusion Geometry