Nonlinear Variable Selection Using Deep Neural Network

• Activity Number: 310 - Topics of Variable Selection
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Yao Chen* and Faming Liang and Xiao Wang
• Companies: Purdue University and Purdue University and Purdue University
• Keywords: variable selection; deep neural network; nonlinear

Abstract

During the past decade, deep learning has demonstrated a great potential in solving many complex artificial intelligence tasks such as pattern recognition and speech understanding. The problems that the current deep learning techniques work well usually have a very large sample size and a relatively small number of features (without or with very few false features). How to adopt deep learning to approximate complex systems with high-dimensional features and make accurate prediction for the system, has posed a great challenge in current research. We have conducted structure selection for the neural network and have developed a novel and efficient algorithm for simultaneous feature selection and prediction. The consistency of variable selection and the error bound of the generalization error are established. The numerical results indicate the proposed method can achieve the state-of-the-art performance on benchmark datasets.