Johnson–Lindenstrauss Lemma

In mathematics, the Johnson–Lindenstrauss lemma is a result named after William B. Johnson and Joram Lindenstrauss concerning low-distortion embeddings of points from high-dimensional into low-dimensional Euclidean space. The lemma states that a set of points in a high-dimensional space can be embedded into a space of much lower dimension in such a way that distances between the points are nearly preserved. The map used for the embedding is at least Lipschitz, and can even be taken to be an orthogonal projection.

The lemma has applications in compressed sensing, manifold learning, dimensionality reduction, and graph embedding. Much of the data stored and manipulated on computers, including text and images, can be represented as points in a high-dimensional space (see vector space model for the case of text). However, the essential algorithms for working with such data tend to become bogged down very quickly as dimension increases. It is therefore desirable to reduce the dimensionality of the data in a way that preserves its relevant structure. The Johnson–Lindenstrauss lemma is a classic result in this vein.

Also, the lemma is tight up to a constant factor, i.e. there exists a set of points of size ''m'' that needs dimension

:$\Omega \left(\frac\right)$

in order to preserve the distances between all pairs of points within a factor of $(1 \pm \varepsilon)$.

Lemma

Given $0 < \varepsilon < 1$, a set $X$ of $m\in\mathbb Z_$ points in $\mathbb^N$ ($N\in\mathbb Z_$), and an integer $n > 8(\ln m)/\varepsilon^2$, there is a linear map $f: \mathbb^N \rightarrow \mathbb^n$ such that

: $(1-\varepsilon)\, u-v\, ^2 \leq \, f(u) - f(v)\, ^2 \leq (1+\varepsilon)\, u-v\, ^2$

for all $u,v \in X$.

The formula can be rearranged:$(1+\varepsilon)^\, f(u)-f(v)\, ^2 \leq \, u-v\, ^2 \leq (1-\varepsilon)^\, f(u)-f(v)\, ^2$

Alternatively, for any $\epsilon\in(0,1)$ and any integer $n\ge15(\ln m)/\varepsilon^2$Or any integer $n>128(\ln m)/(9\varepsilon^2).$ there exists a linear function $f: \mathbb^N \rightarrow \mathbb^n$ such that the restriction $f, _X$ is $(1+\varepsilon)$- bi-Lipschitz.This result follows from the above result. Sketch of proof: Note $1/(1+\varepsilon)<\sqrt$ and $\sqrt<\sqrt<1+\varepsilon$ for all $\varepsilon\in(0,1)$. Do casework for 1=''m'' and 1<''m'', applying the above result to $3\varepsilon/4$ in the latter case, noting $128/9<15.$

One proof of the lemma takes ''ƒ'' to be a suitable multiple of the orthogonal projection onto a random subspace of dimension $n$ in $\mathbb^N$, and exploits the phenomenon of concentration of measure.

An orthogonal projection will, in general, reduce the average distance between points, but the lemma can be viewed as dealing with ''relative distances'', which do not change under scaling. In a nutshell, you roll the dice and obtain a random projection, which will reduce the average distance, and then you scale up the distances so that the average distance returns to its previous value. If you keep rolling the dice, you will, in polynomial random time, find a projection for which the (scaled) distances satisfy the lemma.

Alternate statement

A related lemma is the distributional JL lemma. This lemma states that for any $0 < \varepsilon, \delta < 1/2$ and positive integer $d$, there exists a distribution over $\mathbb^$ from which the matrix $A$ is drawn such that for $k = O(\varepsilon^ \log(1/\delta))$ and for any unit-length vector $x \in \mathbb^$, the claim below holds.
: $P(, \Vert Ax\Vert_2^2-1, >\varepsilon)<\delta$

One can obtain the JL lemma from the distributional version by setting $x = (u-v)/\, u-v\, _2$ and $\delta < 1/n^2$ for some pair ''u'',''v'' both in ''X''. Then the JL lemma follows by a union bound over all such pairs.

Speeding up the JL transform

Given ''A'', computing the matrix vector product takes ''O''(''kd'') time. There has been some work in deriving distributions for which the matrix vector product can be computed in less than ''O''(''kd'') time.

There are two major lines of work. The first, ''Fast Johnson Lindenstrauss Transform'' (FJLT), was introduced by Ailon and Chazelle in 2006.
This method allows the computation of the matrix vector product in just $d\log d + k^$ for any constant $\gamma>0$.

Another approach is to build a distribution supported over matrices that are sparse.
This method allows keeping only an $\varepsilon$ fraction of the entries in the matrix, which means the computation can be done in just $kd\varepsilon$ time.
Furthermore, if the vector has only $b$ non-zero entries, the Sparse JL takes time $kb\varepsilon$, which may be much less than the $d\log d$ time used by Fast JL.

Tensorized random projections

It is possible to combine two JL matrices by taking the so-called face-splitting product, which is defined as the tensor products of the rows (was proposed by V. SlyusarAnna Esteve, Eva Boj & Josep Fortiana (2009): Interaction Terms in Distance-Based Regression, Communications in Statistics - Theory and Methods, 38:19, P. 350

/ref> in 1996 for radar and digital antenna array applications).
More directly, let $\in\mathbb R^$ and $\in\mathbb R^$ be two matrices.
Then the face-splitting product $\bullet$ is
:
\bullet
=
\left \begin
_1 \otimes _1\\\hline
_2 \otimes _2\\\hline
_3 \otimes _3\\
\end
\right

This idea of tensorization was used by Kasiviswanathan et al. 2010Kasiviswanathan, Shiva Prasad, et al. "The price of privately releasing contingency tables and the spectra of random matrices with correlated rows." Proceedings of the forty-second ACM symposium on Theory of computing. 2010. for differential privacy.

JL matrices defined like this use fewer random bits, and can be applied quickly to vectors that have tensor structure, due to the following identity:
:(\mathbf \bull \mathbf)(x\otimes y) = \mathbfx \circ \mathbf y
= \left \begin
(\mathbfx)_1 (\mathbf y)_1 \\
(\mathbfx)_2 (\mathbf y)_2 \\
\vdots
\end\right,
where $\circ$ is the element-wise ( Hadamard) product.
Such computations have been used to efficiently compute polynomial kernels and many other .Woodruff, David P. "Sketching as a Tool for Numerical Linear Algebra." Theoretical Computer Science 10.1-2 (2014): 1-157.

In 2020 it was shown that if the matrices $C_1, C_2, \dots, C_c$ are independent $\pm1$ or Gaussian matrices, the combined matrix $C_1 \bullet \dots \bullet C_c$ satisfies the distributional JL lemma if the number of rows is at least
:$O(\epsilon^\log1/\delta + \epsilon^(\tfrac1c\log1/\delta)^c)$.

For large $\epsilon$ this is as good as the completely random Johnson-Lindenstrauss, but
a matching lower bound in the same paper shows that this exponential dependency on $(\log1/\delta)^c$ is necessary.
Alternative JL constructions are suggested to circumvent this.

See also

* Random projection
*Restricted isometry property

Notes

References

Further reading

*. Journal version of a paper previously appearing at PODC 2001.
*.
*.
*
*
*

{{DEFAULTSORT:Johnson-Lindenstrauss lemma
Lemmas
Metric geometry