Faster Johnson-Lindenstrauss Transforms via Kronecker Products

TL;DR

This paper introduces the Kronecker fast Johnson-Lindenstrauss transform (KFJLT), a method that efficiently embeds vectors with Kronecker structure, significantly reducing computational costs while maintaining embedding quality for high-dimensional data.

Contribution

The paper generalizes the fast Johnson-Lindenstrauss transform to Kronecker-structured data, achieving exponential speedups without substantial loss in embedding accuracy.

Findings

KFJLT reduces embedding cost exponentially compared to FJLT.

High-probability embedding with small distortion for sets in Kronecker-structured spaces.

Application to large-scale Kronecker-structured least squares problems.

Abstract

The Kronecker product is an important matrix operation with a wide range of applications in supporting fast linear transforms, including signal processing, graph theory, quantum computing and deep learning. In this work, we introduce a generalization of the fast Johnson-Lindenstrauss projection for embedding vectors with Kronecker product structure, the Kronecker fast Johnson-Lindenstrauss transform (KFJLT). The KFJLT reduces the embedding cost to an exponential factor of the standard fast Johnson-Lindenstrauss transform (FJLT)'s cost when applied to vectors with Kronecker structure, by avoiding explicitly forming the full Kronecker products. We prove that this computational gain comes with only a small price in embedding power: given $N = \prod_{k=1}^d n_k$, consider a finite set of $p$ points in a tensor product of $d$ constituent Euclidean spaces $\bigotimes_{k=d}^{1}\mathbb{R}^{n_k} \subset \mathbb{R}^{N}$. With high probability, a random KFJLT matrix of dimension $N \times m$ embeds the set of points up to multiplicative distortion $(1\pm \varepsilon)$ provided by $m \gtrsim \varepsilon^{-2} \cdot \log^{2d - 1} (p) \cdot \log N$. We conclude by describing a direct application of the KFJLT to the efficient solution of large-scale Kronecker-structured least squares problems for fitting the CP tensor decomposition.