Spatial-Sign based High dimensional Change Point Inference
Jixuan Liu, Long Feng, Liuhua Peng, Zhaojun Wang · arXiv (2025) · doi:10.48550/arxiv.2504.19306 · gemini-3.1-flash-lite
TL;DR
Traditional methods for detecting sudden changes in data often fail when the data is noisy or has extreme values. The authors developed a new mathematical approach using 'spatial' statistics that is much more reliable for complex, messy data. This new method effectively combines different testing strategies to remain accurate whether changes are concentrated in a few variables or spread across many.
Method and sample: The study utilizes numerical simulation experiments across multivariate normal, Student-t (df=6), and mixture normal distributions. The simulation parameters included N=200 for sample size and varying dimensions P in {100, 200, 300, 400}. Baselines compared against include mean-based methods (CHH, JPYZ, DMS, LZZL, ZWS).
#changepoint-detection#high-dimensional-data#spatial-statistics#robust-inference#asymptotic-independence