Posted on Monday, February 26, 2018
|Date:||Tuesday, Feb 27, 2018|
Abstract Exponential growth in Electronic Healthcare Records (EHR) has resulted in new opportunities for discovery of meaningful data-driven representations and patterns of diseases in Health care research. Deep Learning models have emerged as promising solutions for tackling many health care research problems since they can be automatically trained end-to-end without the need for hand-crafted features. While results from deep learning models are encouraging, there is still a major gap before it can be adopted as the mainstream method for practical healthcare applications. In this talk, I will first discuss the unique challenges that the healthcare data pose for machine learning and data driven analytical systems. Then, I will show the benchmarking performance of the deep learning models with respect to the state-of-the-art machine learning models and prognostic scoring systems on several clinical prediction tasks such as mortality prediction, length of stay prediction, and ICD-9 code group prediction. Finally, I will present our recently proposed novel deep learning models to address the challenges of the healthcare datasets, and discuss one use-case study where our deep learning model was used to learn domain-invariant representations and to transfer knowledge across multi-cohort populations.
Biography Sanjay Purushotham is a Postdoctoral Scholar Research Associate in the Department of Computer Science at the University of Southern California (USC). He obtained his MS and PhD in Electrical Engineering from USC. His research interests are in machine learning, data mining, and its applications to health care & bio-informatics, and data-driven cancer research. Recently, he has been developing deep learning frameworks to model healthcare data and to predict survival outcomes for cancer patients. He has produced more than 25 publications, and won best paper and best runner-up poster awards at the ACM SIGSPATIAL GeoCrowd workshop and SoCal ML Symposium respectively.