Yeseul Jeon, PhD
Postdoctoral Scholar, Biostatistics, TAMU & UCSF
We introduce a Bayesian approach that integrates convolutional neural networks (CNNs) with generalized linear models (GLMs) to enhance both predictive accuracy and statistical inference. By leveraging features from the final CNN layer with Monte Carlo (MC) dropout as covariates in GLMs, our method enables interpretable coefficient estimation and uncertainty quantification. Through ensemble GLMs applied across multiple MC dropout realizations, we account for uncertainties in feature extraction. This approach is applied to malaria incidence, brain tumor imaging, and fMRI data, demonstrating its effectiveness in Bayesian inference for high-dimensional, correlated data in image regression and spatial analysis.