期刊
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
卷 45, 期 7, 页码 9233-9240出版社
IEEE COMPUTER SOC
DOI: 10.1109/TPAMI.2022.3233572
关键词
Training; Feature extraction; Covariance matrices; Linear discriminant analysis; Feedforward neural networks; Eigenvalues and eigenfunctions; Kernel; Fisher's linear discriminant analysis; space-folding operations; rectified linear units
Fisher's linear discriminant analysis (LDA) is a supervised dimensionality reduction method that may be ineffective against complicated class distributions. This paper demonstrates that deep neural networks with rectified linear units can reveal classification information in a subspace where LDA cannot find any. Combining LDA with space-folding operations and fine-tuning can enhance the classification performance. Experimental results on artificial and open data sets validate the feasibility of the proposed approach.
Fisher's linear discriminant analysis (LDA) is an easy-to-use supervised dimensionality reduction method. However, LDA may be ineffective against complicated class distributions. It is well-known that deep feedforward neural networks with rectified linear units as activation functions can map many input neighborhoods to similar outputs by a succession of space-folding operations. This short paper shows that the space-folding operation can reveal to LDA classification information in the subspace where LDA cannot find any. A composition of LDA with the space-folding operation can find classification information more than LDA can do. End-to-end fine-tuning can improve that composition further. Experimental results on artificial and open data sets have shown the feasibility of the proposed approach.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据