Pareto models for discriminative multiclass linear dimensionality reduction
作者：
Highlights：
• We address the class-masking problem (CMP) in Multiclass LDA (MLDA).
• We model MLDA as a set of objective functions for pairwise distances between classes.
• We optimize the compound objective function using Multiobjective Optimization.
• Our models overcome the CMP and are extended to multimodal non-Gaussian data.
• Empirical analysis shows consistent and promising performance in favor of our models.
摘要
Highlights•We address the class-masking problem (CMP) in Multiclass LDA (MLDA).•We model MLDA as a set of objective functions for pairwise distances between classes.•We optimize the compound objective function using Multiobjective Optimization.•Our models overcome the CMP and are extended to multimodal non-Gaussian data.•Empirical analysis shows consistent and promising performance in favor of our models.
论文关键词：Fisher discriminant analysis,Supervised linear dimensionality reduction,Feature transformation,Metric learning,Subspace learning,Multiobjective optimization,Pareto optimality,Kullback–Leibler divergence
论文评审过程：Received 10 January 2014, Revised 13 August 2014, Accepted 8 November 2014, Available online 20 November 2014.
论文官网地址：https://doi.org/10.1016/j.patcog.2014.11.008
