A0317
Title: An estimator for the estimation error and hypothesis testing in non-identifiable models, including machine learning
Authors: Junichiro Yoshida - University of Tokyo, Graduate School of Mathematical Sciences (Japan) [presenting]
Nakahiro Yoshida - University of Tokyo (Japan)
Abstract: To improve the affinity between machine learning and classical statistical methods, it is important to analyze the estimation error (defined as the difference between the expected loss of the estimator and that of the true parameter value) in non-identifiable models, including machine learning. However, in complex non-identifiable models, the estimation error is known to be difficult to calculate specifically. To solve this problem, an estimator is proposed for the estimation error and its confidence interval that can be applied to non-identifiable models, which enables the combination of classical statistical methods, such as hypothesis testing, with machine learning.
