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.

Margin Maximum Embedding Discriminant (Mmed) For Feature Extraction And Classification

Maximum Margin Embedding

Feature extraction using maximum nonparametric margin projection.

Feature Extraction by Maximizing the Average Neighborhood Margin

Unsupervised large margin discriminative projection.

Dimension Reduction by Maximizing Pairwise Discriminations.

Local Maximal Margin Discriminant Embedding for Face Recognition

Feature extraction using graph discriminant embedding

Margin-maximizing feature elimination methods for linear and nonlinear kernel-based discriminant functions

Discriminant Maximum Margin Projections for Face Recognition.

Two Dimensional Maximum Margin Criterion

A Novel Margin Based Algorithm for Feature Extraction

Feature extraction via prototype margin distance maximizing criterion for subspace learning

Data-Dependent Kernel Discriminant Analysis for Feature Extraction and Classification

Geodesic Based Semi-supervised Multi-manifold Feature Extraction

Local Feature Discriminant Projection

Heteroscedastic Max-Min Distance Analysis.

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

Face Recognition by Stepwise Nonparametric Margin Maximum Criterion

Efficient Feature Extraction For Image Classification

Maximum Margin Multi-Dimensional Classification