.png){kind=icon}
.png){kind=icon}
Return
Joint optimization transposed projection envelope linear discriminant analysis mode
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neunet.2026.109475.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Linear Discriminant Analysis (LDA) is a widely employed feature-extraction technique that, guided by Fisher discriminant criterion, projects original high-dimensional samples into a lower-dimensional subspace with enhanced separability. Its interpretability and ease of application are advantageous. However, conventional LDA is constructed at the granularity of the original samples and fails to incorporate correlation information among similar samples, thereby limiting performance. To address this limitation, the present work proposes Transposed Projection Envelope Linear Discriminant Analysis (TPELDA). Through transposed projection, the original samples are transformed into envelope samples that containing correlation information among similar samples. On these envelope samples, Fisher discriminant criterion is applied to learn the dimensionality-reduction subspace, while a distribution discrepancy penalty term ensures the learned subspace remains well suited to the original samples. By jointly optimizing these objectives, TPELDA enhances the discriminative features of samples projected onto the subspace by leveraging the correlation information among similar samples. Experimental results across multiple datasets demonstrate that TPELDA surpasses comparable methods, with average gains in classification accuracy of 2.25 % to 13.19 %. Additional experiments further substantiate the effectiveness of the proposed method.
Keywords:
Linear discriminant analysis
Dimensionality reduction
Correlation information
Distribution discrepancy
Envelope learning
Journal
IF:
6.3
Papers:
7.8K
Citations:
3.0W