This repository provides the codes for the paper "Speed-up and Multi-view extensions to Subclass Discriminant Analysis", preprint available here.

For single-view fastSDA the following files are used:

• calculate_targets_singleview.m - calculates target vector matrix T given class and subclass labels
• get_fastSDA_kernel_results.m - applies kernel fastSDA given data matrix X and labels. Note that X should be sorted according to classes/subclasses
• get_fastSDA_linear_results.m - applies linear fastSDA given the kernel matrix K and labels. Note that data should be sorted according to classes/subclasses

The same logic is followed for multi-view case:

• calculate_target_vectors_multiview.m
• get_fastSDA_multiview_kernel_results.m
• get_fastSDA_multiview_linear_results.m

Additionally, there is now a helper function fastSDA.m to which you can just supply your train data, labels and desired number of dimensions and obtain the projection matrix.

If you find our work useful, please cite it as:

@article{CHUMACHENKO2021107660,
title = "Speed-up and multi-view extensions to subclass discriminant analysis",
journal = "Pattern Recognition",
volume = "111",
pages = "107660",
year = "2021",
issn = "0031-3203",
doi = "https://doi.org/10.1016/j.patcog.2020.107660",
url = "http://www.sciencedirect.com/science/article/pii/S0031320320304635",
author = "Kateryna Chumachenko and Jenni Raitoharju and Alexandros Iosifidis and Moncef Gabbouj"
}