Robust Statistical Distances for Machine Learning

This post discusses the use of statistical distances in comparing distributions and detecting anomalies. Statistical distances are distances between distributions or samples, which can be used in various machine learning applications such as anomaly detection, ordinal regression, and generative adversarial networks (GANs). The article covers visual inspection methods like Q-Q plots and statistical distance measures including the Kolmogorov-Smirnov Distance, Earth Mover's Distance, and Cramér-von Mises Distance. It also highlights the properties of each distance measure and provides an interactive tool for comparing distributions. The post concludes by noting that these methods have their place in different scenarios and can be used effectively to detect anomalies or outliers.