Abstract:We propose a novel discriminative feature learning method via Max-Min Ratio Analysis (MMRA) for exclusively dealing with the long-standing "worst-case class separation" problem. Existing technologies simply consider maximizing the minimal pairwise distance on all class pairs in the low-dimensional subspace, which is unable to separate overlapped classes entirely especially when the distribution of samples within same class is diverging. We propose a new criterion, i.e., Max-Min Ratio Analysis (MMRA) that focuses on maximizing the minimal ratio value of between-class and within-class scatter to extremely enlarge the separability on the overlapped pairwise classes. Furthermore, we develop two novel discriminative feature learning models for dimensionality reduction and metric learning based on our MMRA criterion. However, solving such a non-smooth non-convex max-min ratio problem is challenging. As an important theoretical contribution in this paper, we systematically derive an alternative iterative algorithm based on a general max-min ratio optimization framework to solve a general max-min ratio problem with rigorous proofs of convergence. More importantly, we also present another solver based on bisection search strategy to solve the SDP problem efficiently. To evaluate the effectiveness of proposed methods, we conduct extensive pattern classification and image retrieval experiments on several artificial datasets and real-world ScRNA-seq datasets, and experimental results demonstrate the effectiveness of proposed methods.

Nonparametric Dimension Reduction Via Maximizing Pairwise Separation Probability.

Discriminative Dimension Reduction Via Maximin Separation Probability Analysis

Computational and Theoretical Analysis of Supervised Dimensionality Reduction

Dimension Reduction by Maximizing Pairwise Discriminations.

Dimension Reduction by Minimum Error Minimax Probability Machine

Minimizing Nearest Neighbor Classification Error for Nonparametric Dimension Reduction.

Worst-Case Discriminative Feature Learning via Max-Min Ratio Analysis

Local Feature Discriminant Projection

Divergence Maximizing Linear Projection for Supervised Dimension Reduction

Dimension Reduction with Semi-supervised Pairwise Covariance-Preserving Projection

Supervised locally linear embedding with probability-based distance for classification

Distance metric learning by minimal distance maximization

A Variance Minimization Criterion to Feature Selection Using Laplacian Regularization

Maximin Separation Probability Clustering.

Joint Dimensionality Reduction for Separable Embedding Estimation

Probability-Based Locally Linear Embedding for Classification

Feature extraction using maximum nonparametric margin projection.

A Perception-Driven Approach to Supervised Dimensionality Reduction for Visualization

Neighborhood MinMax Projections

Ultrahigh Dimensional Multi-Class Linear Discriminant Analysis by Pairwise Sure Independence Screening

Discriminant similarity and variance preserving projection for feature extraction