Computing Approximate Statistical Discrepancy

TL;DR

This paper develops algorithms to efficiently approximate the maximum of a function over geometric range spaces with multiple point values, improving accuracy and speed for common geometric shapes.

Contribution

It introduces algorithms for approximating the maximum of a sum-based function over range spaces, with notable improvements for balls, halfspaces, and rectangles.

Findings

Algorithms achieve approximation within epsilon for bounded VC-dimension spaces.

Significant improvements for range spaces defined by balls, halfspaces, and rectangles.

Applicable to discrepancy evaluation, classification, and spatial statistics.

Abstract

Consider a geometric range space $(X,\c{A})$ where each data point $x \in X$ has two or more values (say $r(x)$ and $b(x)$). Also consider a function $\Phi(A)$ defined on any subset $A \in (X,\c{A})$ on the sum of values in that range e.g., $r_A = \sum_{x \in A} r(x)$ and $b_A = \sum_{x \in A} b(x)$. The $\Phi$-maximum range is $A^* = \arg \max_{A \in (X,\c{A})} \Phi(A)$. Our goal is to find some $\hat{A}$ such that $|\Phi(\hat{A}) - \Phi(A^*)| \leq \varepsilon.$ We develop algorithms for this problem for range spaces with bounded VC-dimension, as well as significant improvements for those defined by balls, halfspaces, and axis-aligned rectangles. This problem has many applications in many areas including discrepancy evaluation, classification, and spatial scan statistics.