Abstract:Today, modern databases with "Big Dimensionality" are experiencing a growing trend. Existing approaches that require the calculations of pairwise feature correlations in their algorithmic designs have scored miserably on such databases, since computing the full correlation matrix (i.e., square of dimensionality in size) is computationally very intensive (i.e., million features would translate to trillion correlations). This poses a notable challenge that has received much lesser attention in the field of machine learning and data mining research. Thus, this paper presents a study to fill in this gap. Our findings on several established databases with big dimensionality across a wide spectrum of domains have indicated that an extremely small portion of the feature pairs contributes significantly to the underlying interactions and there exists feature groups that are highly correlated. Inspired by the intriguing observations, we introduce a novel learning approach that exploits the presence of sparse correlations for the efficient identifications of informative and correlated feature groups from big dimensional data that translates to a reduction in complexity from O(m2n) to O(mlogm + Ka mn), where Ka << min(m,n) generally holds. In particular, our proposed approach considers an explicit incorporation of linear and nonlinear correlation measures as constraints in the learning model. An efficient embedded feature selection strategy, designed to filter out the large number of non-contributing correlations that could otherwise confuse the classifier while identifying the correlated and informative feature groups, forms one of the highlights of our approach. We also demonstrated the proposed method on one-class learning, where notable speedup can be observed when solving one-class problem on big dimensional data. Further, to identify robust informative features with minimal sampling bias, our feature selection strategy embeds the V-fold cross validation in the learning model, so as to seek for features that exhibit stable or consistent performance accuracy on multiple data folds. Extensive empirical studies on both synthetic and several real-world datasets comprising up to 30 million dimensions are subsequently conducted to assess and showcase the efficacy of the proposed approach.

CARE: Finding Local Linear Correlations in High Dimensional Data

Finding High-Order Correlations In High-Dimensional Biological Data

Locality Pursuit Embedding

Detecting and Visualizing Profile Correlation in Subspace

Fault Identification Based on Local Feature Correlation

Local Feature Discriminant Projection

SLICE: A Novel Method to Find Local Linear Correlations by Constructing Hyperplanes

Discovering Support and Affiliated Features from Very High Dimensions

Constraint Based Dimension Correlation and Distance Divergence for Clustering High-Dimensional Data

REDUS: Finding Reducible Subspaces in High Dimensional Data.

2D-LPCCA and 2D-SPCCA: Two New Canonical Correlation Methods for Feature Extraction, Fusion and Recognition

Making Trillion Correlations Feasible in Feature Grouping and Selection

Kernel Sparse Locality Preserving Canonical Correlation Analysis for Multi-modal Feature Extraction

Comparative Research on Methods of Dimensionality Reduction in High-Dimension Medical Data

An Information-Theoretic Approach to Unsupervised Feature Selection for High-Dimensional Data

Discriminative Locality Preserving Canonical Correlation Analysis.

Identifying the Complete Correlation Structure in Large-Scale High-Dimensional Data Sets with Local False Discovery Rates

Linear Discriminative Star Coordinates for Exploring Class and Cluster Separation of High Dimensional Data.

Linear Discriminative Star Coordinates for Exploring Class and Cluster Separation of High Dimensional Data

Supervised Discriminative Sparse PCA with Adaptive Neighbors for Dimensionality Reduction

Correlation Structured Low-Rank Subspace Clustering