Dimensionality Reduction and Extremal Parameter Grouping for Weighted Graph Problems

Abstract

This work presents graph cut approach to dimensionality reduction and shows connections between the problem of extremal parameter grouping and the problem of maximization of positive definite quadratic form on the set of vertices of k-dimensional hypercube of side 1 by modulo. It is shown that the last problem is reduced to the minimum negative cut search for a weighted graph. A heuristic algorithm for solving this problem is given.