报告题目： Ditributionally Robust Learning and Applications to Predictive and Prescriptive Health Analytics

报告人：Yannis Paschalidis, Boston University

报告时间：2018年7月17日 上午10:00—11:00

报告地点：南一楼中311

报告摘要：

We present a distributionally robust optimization approach to learning predictive models, either in the context of classification or regression. Motivated by medical applications, we assume that training data are contaminated with (unknown) outliers, which have the effect of skewing the parameters of the model. Our robust learning approach is able to guard against such outliers and learn model parameters consistent with the non-outlying data. We establish rigorous out-of-sample guarantees on the performance of the method and develop extensions to nonlinear models, where different predictive models are used for different clusters of the data, or even individual data points.

Beyond predictions, we devise methods that can leverage the robust predictive models to make decisions and offer specific personalized prescriptions and recommendations to improve future outcomes. We will discuss several medical applications of our methods, including predicting hospitalizations for chronic disease patients, predicting and preventing re-admissions following general surgery, predicting hospital length-of-stay for surgical patients, and detecting CT scans with an abnormal radiation dosage delivered to the patient.

报告人简介：

Yannis Paschalidis is a Professor in the College of Engineering at Boston University with joint appointments in the Department of Electrical and Computer Engineering, the Division of Systems Engineering, and the Department of Biomedical Engineering.

He is the Director of the Center for Information and Systems Engineering (CISE) – a Boston University research center with 39 affiliated faculty and more than $6.6 million of annual research expenditures.

He is also affiliated with the BioMolecular Engineering Research Center (BMERC), the Clinical & Translational Science Institute (CTSI), the Precision Diagnostics Center (PDC), and the Rafik B. Hariri Institute for Computing and Computational Science & Engineering.