A Martingale Kernel Independence Test
Researchers have developed a new statistical test called the Martingale Kernel Independence Test (mHSIC and mdHSIC) to efficiently assess the independence of variables. This new method offers a significant speedup, running 25 to 60 times faster than existing permutation-based tests by replacing computationally intensive permutation steps with a single normal-quantile lookup. The mHSIC statistic achieves quadratic cost and is consistent against all fixed alternatives, while the mdHSIC statistic offers finite-sample consistency with a linear cost in the number of tested variables. AI
IMPACT Introduces a faster statistical test for variable independence, potentially accelerating research and model development that relies on such analyses.