Supervised dimensionality reduction via distance correlation maximization

DOI10.1214/18-EJS1403zbMATH Open1390.62105arXiv1601.00236OpenAlexW2963297095WikidataQ130149194 ScholiaQ130149194MaRDI QIDQ1746548FDOQ1746548

Authors: Praneeth Vepakomma, Chetan Tonde, Ahmed Elgammal 

Publication date: 25 April 2018

Published in: Electronic Journal of Statistics (Search for Journal in Brave)

Abstract: In our work, we propose a novel formulation for supervised dimensionality reduction based on a nonlinear dependency criterion called Statistical Distance Correlation, Szekely et. al. (2007). We propose an objective which is free of distributional assumptions on regression variables and regression model assumptions. Our proposed formulation is based on learning a low-dimensional feature representation $mathbfz$, which maximizes the squared sum of Distance Correlations between low dimensional features $mathbfz$ and response $y$, and also between features $mathbfz$ and covariates $mathbfx$. We propose a novel algorithm to optimize our proposed objective using the Generalized Minimization Maximizaiton method of Parizi et. al. (2015). We show superior empirical results on multiple datasets proving the effectiveness of our proposed approach over several relevant state-of-the-art supervised dimensionality reduction methods.

Full work available at URL: https://arxiv.org/abs/1601.00236

Recommendations

• scientific article; zbMATH DE number 5957198
• Sufficient dimension reduction via squared-loss mutual information estimation
• Dependence maps, a dimensionality reduction with dependence distance for high-dimensional data
• Supervised principal component analysis: visualization, classification and regression on subspaces and submanifolds
• Large correlation analysis

zbMATH Keywords

optimizationdistance correlationfixed point iterationrepresentation learningsupervised dimensionality reductionminorization maximizationmultivariate statistical independence

Mathematics Subject Classification ID

Nonparametric regression and quantile regression (62G08) Asymptotic properties of nonparametric inference (62G20) Measures of association (correlation, canonical correlation, etc.) (62H20)

Cites Work

• Measuring and testing dependence by correlation of distances
• Variable selection in functional data classification: a maxima-hunting proposal
• Modern multidimensional scaling. Theory and applications.
• Sliced Inverse Regression for Dimension Reduction
• Feature screening via distance correlation learning
• Title not available (Why is that?)
• Hedonic housing prices and the demand for clean air
• The distance correlation \(t\)-test of independence in high dimension
• On Nonlinear Fractional Programming
• Optimization
• Brownian distance covariance
• Multidimensional scaling. I: Theory and method
• Gradient-based kernel dimension reduction for regression
• Likelihood-based sufficient dimension reduction
• Marginal tests with sliced average variance estimation
• Title not available (Why is that?)
• Graphics for Regressions With a Binary Response
• Density ratio estimation in machine learning. Foreword by Thomas G. Dietterich
• On the uniqueness of distance covariance
• Sequential Minimax Search for a Maximum
• Sufficient dimension reduction via squared-loss mutual information estimation
• The sliced inverse regression algorithm as a maximum likelihood procedure
• Sliced inverse regression for multivariate response regression
• Minimization of ratios
• Partial central subspace and sliced average variance estimation
• On convergence of minimization methods: Attraction, repulsion, and selection
• Diagnostic studies in sufficient dimension reduction
• Constructive setting for problems of density ratio estimation

Cited In (15)

• Direct estimation of the derivative of quadratic mutual information with application in supervised dimension reduction
• Large correlation analysis
• A dimensionality reduction method of continuous dependent variables based supervised Laplacian eigenmaps
• Feature selection based on distance correlation: a filter algorithm
• Supervised t-Distributed Stochastic Neighbor Embedding for Data Visualization and Classification
• Supervised dimension reduction of intrinsically low-dimensional data
• Title not available (Why is that?)
• Dimensionality reduction using automatic supervision for vision-based terrain learning
• Unsupervised dimensionality reduction versus supervised regularization for classification from sparse data
• Dependence maps, a dimensionality reduction with dependence distance for high-dimensional data
• Asymptotic distributions of high-dimensional distance correlation inference
• Supervised distance preserving projection using alternating direction method of multipliers
• Correlation-based multidimensional scaling for unsupervised subspace learning
• Diverse data selection via combinatorial quasi-concavity of distance covariance: a polynomial time global minimax algorithm
• Convergence rates for kernel regression in infinite-dimensional spaces

Uses Software

• UCI-ml

This page was built for publication: Supervised dimensionality reduction via distance correlation maximization

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1746548)