Online Change Point Detection for Regression Coefficients via weighted Empirical Risk Minimization
-
SeriesResearch Master Defense
-
Speaker
-
LocationVrije Universiteit Amsterdam, room HG-01A43
Amsterdam -
Date and time
July 07, 2026
15:00 - 17:00
This thesis studies online change-point detection for changes in linear regression coefficients. The goal is to detect, as new data arrive, whether the relationship between covariates and the response variable has changed. The proposed method is based on weighted empirical risk minimisation. At each step, the procedure looks at a recent window of observations and compares a one-regime regression model with a two-regime model that allows for a possible change point. If the two-regime model improves the fit sufficiently, the procedure raises an alarm.