|
Abstract:
|
Regression models with varying coefficients have been widely used to study the associations between the massive imaging data and variables of interests. This problem is challenging, due to the ultrahigh dimensionality, the high and heterogeneous level of noise, and the limited sample size of the imaging data. In this talk, I will present a series of varying coefficient models for imaging data analysis where the varying coefficients are constructed through deep neural networks (DNN). Compared with the existing solutions, our methods are more flexible in capturing the complex patterns among the brain activity, of which the noise level and the spatial dependence appear to be heterogeneous across different brain regions. I will discuss the parameter estimation and statistical inference procedures along with the theoretical properties of our proposed methods. I will show that the new methods outperform the existing ones through both simulations and different neuroimaging data examples.
|
