Abstract
Parametric Approximation Algorithms for High-Dimensional Euclidean Similarity
by: Omer Egecioglu
Abstract:
We introduce a spectrum of algorithms for measuring the similarity of high-dimensional vectors in Euclidean space. The algorithms proposed consist of a convex combination of two measures: one which contains summary data about the shape of a vector, and the other about the relative magnitudes of the coordinates. We present experiments on time-series data on labor statistics unemploymentfigures that show the effectiveness of the algorithm as a function of the parameter that combines the two parts.
Keywords:
Dimensionality reduction, similarity search, inner-product, symmetric function, permutation, inversion, time-series data, labor statistics.
Date:
July 2000
Document: 2000-14
