A schematic two-dimensional cloud and a model supported on a line. The paper represents dimension reduction through tempered distributions, using a penalty to encourage low-dimensional support and an alternating scheme to find the model. Research source
About this paper
Reformulates dimension reduction through tempered distributions supported on low-dimensional subspaces.
The formulation connects unsupervised and sufficient dimension reduction. A penalty encourages low-dimensional support, leading to an alternating optimization method and a practical approximation based on maximum mean discrepancy.
Citation
BibTeX
@article{takhanov2023dimensionreductiondistributions,
title = {Reducing the dimensionality of data using tempered distributions},
author = {Takhanov, Rustem},
year = {2023},
journal = {Digital Signal Processing},
eprint = {1903.05083},
archivePrefix = {arXiv},
url = {https://arxiv.org/pdf/1903.05083}
}