Incremental Sparse Binary Vector Similarity Search in High-Dimensional
Given a sparse binary matrix A and a sparse query vector x, can we efficiently identify the large entries of the matrix-vector product Ax? This problem occurs in document comparison, spam filtering, network intrusion detection, information retrieval, as well as other areas. We present an exact deterministic algorithm that takes advantage of the sparseness of A and x. Although in the worst case, the query complexity is linear in the number of rows of A, the amortized query complexity for a sequence of several similar queries depends only logarithmically on the size of A when the non-zero entries of A and the queries are distributed uniformly.
Pre-2018 CSE ID: CS2006-0866