The emergence and availability of electronic health records can enable disease prevention, streamline clinical decision-making, and improve diagnostic precision. However, one major challenge is how to produce appropriate representations from these high-dimensional data. Tensors, generalizations of matrices to multiway data, are natural structures for capturing higher-order interactions. Factorization of these tensors provides a powerful, data-driven framework for learning representations across a variety of downstream prediction tasks. Unfortunately, tensor factorization is a computationally expensive task. In this talk, I will present our work on how to accelerate convergence with limited computing resources and how to jointly learn representations across distributed locations while preserving the privacy of patient information.
Bio: Joyce Ho is an Associate Professor in the Computer Science Department at Emory University. She received my Ph.D. in Electrical and Computer Engineering from the University of Texas at Austin, and an M.A. and B.S. in Electrical Engineering and Computer Science from Massachusetts Institute of Technology. Her research focuses on the development of novel machine learning algorithms to address problems in healthcare such as identifying patient subgroups or phenotypes, integration of new streams of data, fusing different modalities of data (e.g., structured medical codes and unstructured text), and dealing with conflicting expert annotations. Her work has been supported by the National Science Foundation (including a CAREER award), National Institute of Health, Robert Wood Johnson Foundation, and Johnson and Johnson.
See all upcoming talks at https://www.anl.gov/mcs/lans-seminars