Identifying nonlinear variation patterns with deep autoencoders

Abstract

The discovery of nonlinear variation patterns in high-dimensional profile data is an important task in many quality control and manufacturing settings. We present an automated method for discovering nonlinear variation patterns using deep autoencoders. The approach provides a functional mapping from a low-dimensional representation to the original spatially-dense feature space of the profile data that is both interpretable and efficient with respect to preserving information. We compare our deep autoencoder approach to several other methods for discovering variation patterns in profile data. Our results indicate that deep autoencoders consistently outperform the alternative approaches in reproducing the original profiles from the learned variation sources.

KEYWORDS:

• Profile data
• variation pattern
• autoencoder
• autoassociative neural network
• deep learning
• visualization

Additional information

Notes on contributors

Phillip Howard

Phillip Howard received a Ph.D. in industrial engineering from Arizona State University in 2016 with a research emphasis in machine learning. He is currently a Data Scientist at Intel Corporation where he focuses on enabling supply chain intelligence & analytics capabilities using statistics and machine learning methods. His research interests include applications of deep learning for nonlinear dimensionality reduction, cognitive computing solutions for supply chain monitoring, and interpretable feature learning from heterogeneous data sources in health care applications.

Daniel W. Apley

Daniel W. Apley is a professor of industrial engineering and management sciences at Northwestern University, Evanston, IL. He obtained B.S., M.S., and Ph.D. degrees in mechanical engineering and an M.S. degree in electrical engineering from the University of Michigan. His research interests are at the interface of engineering modeling, statistical analysis, and predictive analytics, with particular emphasis on enterprise process modeling and manufacturing variation reduction applications in which large amounts of data are available. His research has been supported by numerous industries and government agencies. He received the NSF CAREER award in 2001, the IIE Transactions Best Paper Award in 2003, and the Wilcoxon Prize for best practical application paper appearing in Technometrics in 2008. He is Editor-in-Chief of Technometrics and has served as Editor-in-Chief for the Journal of Quality Technology, Chair of the Quality, Statistics & Reliability Section of INFORMS, and Director of the Manufacturing and Design Engineering Program at Northwestern.

George Runger

George Runger is the chair of the Department of Biomedical Informatics and professor in the School of Computing, Informatics, and Decision Systems Engineering at Arizona State University. He was an affiliated faculty member with BMI and the related Center for Health Information & Research for several years. He researches analytical methods for knowledge generation and data-driven improvements in organizations. He focuses on machine learning for large, complex data, and real-time analysis, with applications to surveillance, decision support, and population health. Previously, he was a senior engineer and technical leader for data analytics projects at IBM. He holds degrees in industrial engineering and statistics. He has over 100 publications in research journals with funding from federal and corporate sponsors. He reviews for many journals in the area of machine learning and statistics and he is currently the department editor for healthcare informatics for IISE Transactions on Healthcare Systems Engineering.