主 题: Cross validation for comparing learning procedures
报告人: Prof.Yuhong Yang (University of Minnesota)
时 间: 2006-07-06 下午 3:30 - 4:30
地 点: 理科一号楼 1114
Cross validation (CV) is a general tool for comparing statistical learning
procedures. Shao (1997) discovered the surprising fact that for comparing
linear regression models, the size of the evaluation part in the data
splitting has to be dominating in order for CV to choose the true model
consistently. What happens when we compare two general learning procedures
(e.g., one parametric and one nonparametric, or two nonparametric
procedures)? We show that the situation can be completely different. We give
sufficient conditions to sure consistency in selection in the contexts of
regression and classification.
