Sparse Principal Component Analysis via Variable Projection

Abstract

Sparse principal component analysis (SPCA) has emerged as a powerful technique for modern data analysis, providing improved interpretation of low-rank structures by identifying localized spatial structures in the data and disambiguating between distinct time scales. We demonstrate a robust and scalable SPCA algorithm by formulating it as a value-function optimization problem. This viewpoint leads to a flexible and computationally efficient algorithm. Further, we can leverage randomized methods from linear algebra to extend the approach to the large-scale (big data) setting. Our proposed innovation also allows for a robust SPCA formulation which obtains meaningful sparse principal components in spite of grossly corrupted input data. The proposed algorithms are demonstrated using both synthetic and real world data, and show exceptional computational efficiency and diagnostic performance.

References in corpus (28)

• Discussion of "Least angle regression" by Efron et al
• Rejoinder to "Least angle regression" by Efron et al
• Discussion of "Least angle regression" by Efron et al
• Least Angle Regression
• Discussion of "Least angle regression" by Efron et al
• Discussion of "Least angle regression" by Efron et al
• Discussion of "Least angle regression" by Efron et al
• Discussion of "Least angle regression" by Efron et al
• Discussion of "Least angle regression" by Efron et al
• Discussion of "Least angle regression" by Efron et al
• Discovering governing equations from data: Sparse identification of nonlinear dynamical systems
• Data-Driven Sparse Sensor Placement for Reconstruction
• Generalized power method for sparse principal component analysis
• Linear Dimensionality Reduction: Survey, Insights, and Generalizations
• Sparse PCA: Optimal rates and adaptive estimation
• Cluster-based reduced-order modelling of a mixing layer
• Decomposition into Low-rank plus Additive Matrices for Background/Foreground Separation: A Review for a Comparative Evaluation with a Large-Scale Dataset
• Optimal Solutions for Sparse Principal Component Analysis
• Compressed Modes for Variational Problems in Mathematics and Physics
• Sparse Dynamics for Partial Differential Equations
• Adaptive $h$-refinement for reduced-order models
• A direct formulation for sparse PCA using semidefinite programming
• Network Structure of Two-Dimensional Decaying Isotropic Turbulence
• Randomized Matrix Decompositions using R
• Estimating Nuisance Parameters in Inverse Problems
• A minimization principle for the description of time-dependent modes associated with transient instabilities
• Dynamical indicators for the prediction of bursting phenomena in high-dimensional systems
• Why are Big Data Matrices Approximately Low Rank?

Cited by in corpus (13)

• Shallow Neural Networks for Fluid Flow Reconstruction with Limited Sensors
• Randomized Nonnegative Matrix Factorization
• Numerical Approximation of Partial Differential Equations by a Variable Projection Method with Artificial Neural Networks
• A generalization of regularized dual averaging and its dynamics
• NEU: A Meta-Algorithm for Universal UAP-Invariant Feature Representation
• An Alternating Manifold Proximal Gradient Method for Sparse PCA and Sparse CCA
• Sparse principal component regression via singular value decomposition approach
• Improving full-waveform inversion based on sparse regularisation for geophysical data
• Error Estimation for Sketched SVD via the Bootstrap
• Analysis of Truncated Orthogonal Iteration for Sparse Eigenvector Problems
• Interpretable Deep Learning for Stock Returns: A Consensus-Bottleneck Asset Pricing Model
• RetinaMatch: Efficient Template Matching of Retina Images for Teleophthalmology
• Comparison of Clustering Methods for Extraction of Uncorrelated Sparse Sources from Data Mixtures