Abstract:Linear discriminant analysis (LDA) represents a simple yet powerful technique for partitioning a p-dimensional feature vector into one of K classes based on a linear projection learned from N labeled observations. However, it is well-established that in the high-dimensional setting (p > N) the underlying projection estimator degenerates. Moreover, any linear discriminate function involving a large number of features may be difficult to interpret. To ameliorate these issues, two general categories of sparse LDA modifications have been proposed, both to reduce the number of active features and to stabilize the resulting projections. The first, based on optimal scoring, is more straightforward to implement and analyze but has been heavily criticized for its ambiguous connection with the original LDA formulation. In contrast, a second strategy applies sparse penalty functions directly to the original LDA objective but requires additional heuristic trade-off parameters, has unknown global and local minima properties, and requires a greedy sequential optimization procedure. In all cases the choice of sparse regularizer can be important, but no rigorous guidelines have been provided regarding which penalty might be preferable. Against this backdrop, we winnow down the broad space of candidate sparse LDA algorithms and promote a specific selection based on optimal scoring coupled with a particular, complementary sparse regularizer. This overall process ultimately progresses our understanding of sparse LDA in general, while leading to targeted modifications of existing algorithms that produce superior results in practice on three high-dimensional gene data sets.

Theoretical Analysis of Direct LDA in Small Sample Size Problem

Discriminative analysis of direct linear discriminant analysis for face recognition

Computational and Theoretical Analysis of Supervised Dimensionality Reduction

Linear discriminant analysis for the small sample size problem: an overview

An Improved DLDA Based Method- Nonparametric DLDA

Two Dimension Linear Discriminant Analysis Based on Image Sampling and Regroupment

A new LDA-based face recognition system which can solve the small sample size problem

From NLDA to LDA/GSVD: a Modified NLDA Algorithm

Constrained Linear Discrimination Analysis and Face Recognition

Neighborhood linear discriminant analysis

Understanding and Evaluating Sparse Linear Discriminant Analysis

Optimal Feature Selection for Sparse Linear Discriminant Analysis and Its Applications in Gene Expression Data

A Discriminant Model for the Pattern Recognition of Linearly Independent Samples

Making Fisher Discriminant Analysis Scalable.

Least Square Incremental Linear Discriminant Analysis

Least squares online linear discriminant analysis

Robust Linear Discriminant Analysis Using Ratio Minimization of L1,2-Norms

Sparse linear discriminant analysis by thresholding for high dimensional data

3E-LDA

On the Equivalence of Linear Discriminant Analysis and Least Squares Regression

Data-Dependent Kernel Discriminant Analysis for Feature Extraction and Classification