A0514
Title: Prediction intervals with high dimensional models: With applications in LASSO and deep neural networks
Authors: Zhe Fei - UC Riverside (United States) [presenting]
Abstract: A new approach is presented for making inferences about the prediction of continuous outcomes in high-dimensional settings where the number of features greatly exceeds the number of observations. Our method involves repeated applications of the Lasso procedure on the resampled data, treating the resulting smooth learner as a U-statistic. The theoretical properties of the smooth learner are established, and a consistent variance estimator is developed to quantify prediction uncertainty. Our approach is extended to deep neural networks, and the prediction intervals are derived. To demonstrate the effectiveness of our method, simulations have been conducted and applied them to a real-world dataset to predict the DNA methylation age of patients with different tissue samples, which may adequately characterize the ageing process.
