Novel LSH scheme improves running time for approximate nearest neighbor problems, indicating significant efficiency gains.
Key Points
The research aims to develop an efficient Locality-Sensitive Hashing (LSH) scheme for solving the Approximate Nearest Neighbor problem using p-stable distributions.
Introduced a novel LSH scheme based on p-stable distributions for different lp norms.
Developed exact neighbor finding in O(log n) time under a bounded growth condition.
Conducted experiments on synthetic data sets to compare performance with kd-tree.
The proposed LSH scheme demonstrated up to 40 times faster performance than kd-tree on synthetic datasets.
Achieved the first provably efficient approximate NN algorithm for cases where p<1.
The resulting query time bounds are simplified and free from large factors.