An Information-Theoretic Measure of Dependency Among Variables in Large Datasets

TL;DR

This paper introduces a computationally efficient approximation to the maximal information coefficient (MIC) for measuring dependence between variables, enabling scalable analysis of large datasets while maintaining detection of linear and non-linear associations.

Contribution

It proposes a new approximation method for MIC using uniform data partitioning, reducing computational cost significantly.

Findings

The approximation closely matches the original MIC in detecting dependencies.

Experiments show the method is faster and scalable for large datasets.

The approach maintains high accuracy in various dependency scenarios.

Abstract

The maximal information coefficient (MIC), which measures the amount of dependence between two variables, is able to detect both linear and non-linear associations. However, computational cost grows rapidly as a function of the dataset size. In this paper, we develop a computationally efficient approximation to the MIC that replaces its dynamic programming step with a much simpler technique based on the uniform partitioning of data grid. A variety of experiments demonstrate the quality of our approximation.