Central Limit Theorem for Blurring Process


Ting-Li Chen

13:30:00 - 15:00:00

103 , Mathematics Research Center Building (ori. New Math. Bldg.)

Many statistical methods of estimation obtain their solutions numerically through iterative processes. Currently, most iterative methods are based on nonblurring processes, where the estimation is iteratively updated to improve the fit to the original data. An alternative approach is to update both the estimation and data iteratively. Such a process is considered as a blurring process. Because the mechanism is more complicated, the properties of blurring process are less studied in the literature. In our previous work, we proved the convergence and the consistency of blurring process. In this talk, I will discuss a Central Limit Theorem for blurring process. I will also show that blurring process is more efficient than nonblurring process in our simulation studies.