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.

Discriminative Dimension Reduction Via Maximin Separation Probability Analysis

Dimension Reduction by Maximizing Pairwise Discriminations.

Computational and Theoretical Analysis of Supervised Dimensionality Reduction

Nonparametric Dimension Reduction Via Maximizing Pairwise Separation Probability.

Dimension Reduction by Minimum Error Minimax Probability Machine

Maximin Separation Probability Clustering.

Max-Min Distance Analysis by Using Sequential SDP Relaxation for Dimension Reduction

Distance metric learning by minimal distance maximization

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

A Perception-Driven Approach to Supervised Dimensionality Reduction for Visualization

Local Feature Discriminant Projection

Data-Dependent Kernel Discriminant Analysis for Feature Extraction and Classification

Divergence Maximizing Linear Projection for Supervised Dimension Reduction

Multimodal Linear Discriminant Analysis Via Structural Sparsity

Quadratic Discriminant Analysis by Projection

Discriminant Maximum Margin Projections for Face Recognition.

A New Reduced-Rank Linear Discriminant Analysis Method and Its Applications

Supervised Dimensionality Reduction Technology of Generalized Discriminant Component Analysis and Its Kernelization Forms

Quantum Dimensionality Reduction by Linear Discriminant Analysis

Nonparametric Discriminant Multi-Manifold Learning for Dimensionality Reduction

Minimizing Nearest Neighbor Classification Error for Nonparametric Dimension Reduction.