The Fast Johnson-Lindenstrauss Transform Is Even Faster

Ora Nova Fandina, Mikael Møller Høgsgaard, Kasper Green Larsen
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:9689-9715, 2023.

Abstract

The Johnson-Lindenstaruss lemma (Johnson & Lindenstrauss, 1984) is a cornerstone result in dimensionality reduction, stating it is possible to embed a set of $n$ points in $d$-dimensional Euclidean space into optimal $k=O(\varepsilon^{-2} \ln n)$ dimensions, while preserving all pairwise distances to within a factor $(1 \pm \varepsilon)$. The seminal Fast Johnson-Lindenstrauss (Fast JL) transform by Ailon and Chazelle (SICOMP’09) supports computing the embedding of a data point in $O(d \ln d +k \ln^2 n)$ time, where the $d \ln d$ term comes from multiplication with a $d \times d$ Hadamard matrix and the $k \ln^2 n$ term comes from multiplication with a sparse $k \times d$ matrix. Despite the Fast JL transform being more than a decade old, it is one of the fastest dimensionality reduction techniques for many tradeoffs between $\varepsilon, d$ and $n$. In this work, we give a surprising new analysis of the Fast JL transform, showing that the $k \ln^2 n$ term in the embedding time can be improved to $(k \ln^2 n)/\alpha$ for an $\alpha = \Omega(\min\{\varepsilon^{-1}\ln(1/\varepsilon), \ln n\})$. The improvement follows by using an even sparser matrix. We complement our improved analysis with a lower bound showing that our new analysis is in fact tight.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-fandina23a, title = {The Fast Johnson-Lindenstrauss Transform Is Even Faster}, author = {Fandina, Ora Nova and H{\o}gsgaard, Mikael M{\o}ller and Larsen, Kasper Green}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {9689--9715}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/fandina23a/fandina23a.pdf}, url = {https://proceedings.mlr.press/v202/fandina23a.html}, abstract = {The Johnson-Lindenstaruss lemma (Johnson & Lindenstrauss, 1984) is a cornerstone result in dimensionality reduction, stating it is possible to embed a set of $n$ points in $d$-dimensional Euclidean space into optimal $k=O(\varepsilon^{-2} \ln n)$ dimensions, while preserving all pairwise distances to within a factor $(1 \pm \varepsilon)$. The seminal Fast Johnson-Lindenstrauss (Fast JL) transform by Ailon and Chazelle (SICOMP’09) supports computing the embedding of a data point in $O(d \ln d +k \ln^2 n)$ time, where the $d \ln d$ term comes from multiplication with a $d \times d$ Hadamard matrix and the $k \ln^2 n$ term comes from multiplication with a sparse $k \times d$ matrix. Despite the Fast JL transform being more than a decade old, it is one of the fastest dimensionality reduction techniques for many tradeoffs between $\varepsilon, d$ and $n$. In this work, we give a surprising new analysis of the Fast JL transform, showing that the $k \ln^2 n$ term in the embedding time can be improved to $(k \ln^2 n)/\alpha$ for an $\alpha = \Omega(\min\{\varepsilon^{-1}\ln(1/\varepsilon), \ln n\})$. The improvement follows by using an even sparser matrix. We complement our improved analysis with a lower bound showing that our new analysis is in fact tight.} }
Endnote
%0 Conference Paper %T The Fast Johnson-Lindenstrauss Transform Is Even Faster %A Ora Nova Fandina %A Mikael Møller Høgsgaard %A Kasper Green Larsen %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-fandina23a %I PMLR %P 9689--9715 %U https://proceedings.mlr.press/v202/fandina23a.html %V 202 %X The Johnson-Lindenstaruss lemma (Johnson & Lindenstrauss, 1984) is a cornerstone result in dimensionality reduction, stating it is possible to embed a set of $n$ points in $d$-dimensional Euclidean space into optimal $k=O(\varepsilon^{-2} \ln n)$ dimensions, while preserving all pairwise distances to within a factor $(1 \pm \varepsilon)$. The seminal Fast Johnson-Lindenstrauss (Fast JL) transform by Ailon and Chazelle (SICOMP’09) supports computing the embedding of a data point in $O(d \ln d +k \ln^2 n)$ time, where the $d \ln d$ term comes from multiplication with a $d \times d$ Hadamard matrix and the $k \ln^2 n$ term comes from multiplication with a sparse $k \times d$ matrix. Despite the Fast JL transform being more than a decade old, it is one of the fastest dimensionality reduction techniques for many tradeoffs between $\varepsilon, d$ and $n$. In this work, we give a surprising new analysis of the Fast JL transform, showing that the $k \ln^2 n$ term in the embedding time can be improved to $(k \ln^2 n)/\alpha$ for an $\alpha = \Omega(\min\{\varepsilon^{-1}\ln(1/\varepsilon), \ln n\})$. The improvement follows by using an even sparser matrix. We complement our improved analysis with a lower bound showing that our new analysis is in fact tight.
APA
Fandina, O.N., Høgsgaard, M.M. & Larsen, K.G.. (2023). The Fast Johnson-Lindenstrauss Transform Is Even Faster. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:9689-9715 Available from https://proceedings.mlr.press/v202/fandina23a.html.

Related Material