Abstract:We present a new manifold learning algorithm called Local Orthogonality Preserving Alignment (LOPA). Our algorithm is inspired by the Local Tangent Space Alignment (LTSA) method that aims to align multiple local neighborhoods into a global coordinate system using affine transformations. However, LTSA often fails to preserve original geometric quantities such as distances and angles. Although an iterative alignment procedure for preserving orthogonality was suggested by the authors of LTSA, neither the corresponding initialization nor the experiments were given. Procrustes Subspaces Alignment (PSA) implements the orthogonality preserving idea by estimating each rotation transformation separately with simulated annealing. However, the optimization in PSA is complicated and multiple separated local rotations may produce globally contradictive results. To address these difficulties, we first use the pseudo-inverse trick of LTSA to represent each local orthogonal transformation with the unified global coordinates. Second the orthogonality constraints are relaxed to be an instance of semi-definite programming (SDP). Finally a two-step iterative procedure is employed to further reduce the errors in orthogonal constraints. Extensive experiments show that LOPA can faithfully preserve distances, angles, inner products, and neighborhoods of the original datasets. In comparison, the embedding performance of LOPA is better than that of PSA and comparable to that of state-of-the-art algorithms like MVU and MVE, while the runtime of LOPA is significantly faster than that of PSA, MVU and MVE.

Nonlinear Dimensionality Reduction with Local Spline Embedding

Spline embedding for nonlinear dimensionality reduction

Image Feature Optimization Based on Nonlinear Dimensionality Reduction

Deep Manifold Transformation for Dimension Reduction

Locality Pursuit Embedding

Deep Manifold Transformation for Nonlinear Dimensionality Reduction

Nonlinear Dimension Reduction Via Local Tangent Space Alignment

Deep Manifold Computing and Visualization Using Elastic Locally Isometric Smoothness

Deep Manifold Computing and Visualization

Stable Local Dimensionality Reduction Approaches

Distortion-Free Nonlinear Dimensionality Reduction

Principal Manifolds and Nonlinear Dimensionality Reduction Via Tangent Space Alignment

Nonlinear Dimensionality Reduction by Local Orthogonality Preserving Alignment

Principal manifolds and nonlinear dimension reduction via local tangent space alignment (dept of computer science, pennsylvania state univ, university park, pa)

Local Feature Discriminant Projection

Maximal Linear Embedding for Dimensionality Reduction

Adaptive Nonlinear Dimensionality Reduction with a Local Metric.

Riemannian manifold learning for nonlinear dimensionality reduction

Nonlinear Supervised Dimensionality Reduction via Smooth Regular Embeddings

Nonlinear embedding preserving multiple local-linearities

Local Regularized Least-Square Dimensionality Reduction