Optimal Las Vegas Locality Sensitive Data Structures

Ahle Thomas Dybdahl. Arxiv 2017

[Paper]    
ARXIV LSH Unsupervised

We show that approximate similarity (near neighbour) search can be solved in high dimensions with performance matching state of the art (data independent) Locality Sensitive Hashing, but with a guarantee of no false negatives. Specifically, we give two data structures for common problems. For \(c\)-approximate near neighbour in Hamming space we get query time \(dn^{1/c+o(1)}\) and space \(dn^{1+1/c+o(1)}\) matching that of \cite{indyk1998approximate} and answering a long standing open question from~\cite{indyk2000dimensionality} and~\cite{pagh2016locality} in the affirmative. By means of a new deterministic reduction from \(\ell_1\) to Hamming we also solve \(\ell_1\) and \(ℓ₂\) with query time \(d^2n^{1/c+o(1)}\) and space \(d^2 n^{1+1/c+o(1)}\). For \((s_1,s_2)\)-approximate Jaccard similarity we get query time \(dn^{\rho+o(1)}\) and space \(dn^{1+\rho+o(1)}\), \(\rho=log\frac{1+s_1}{2s_1}\big/log\frac{1+s_2}{2s_2}\), when sets have equal size, matching the performance of~\cite{tobias2016}. The algorithms are based on space partitions, as with classic LSH, but we construct these using a combination of brute force, tensoring, perfect hashing and splitter functions `a la~\cite{naor1995splitters}. We also show a new dimensionality reduction lemma with 1-sided error.

Similar Work