CSE DSI Machine Learning Seminar with Aleksandr Aravkin (Applied Math, U Washington)

Fusing parametric and nonparametric estimation to obtain global health results using heterogeneous data

Key estimates in global health, including estimates of all-cause mortality by location, sex, age, and year, are difficult to compute because of data heterogeneity. Many high-income locations, such as the US, have high quality data, while in other locations, data sources are very noisy or entirely absent. To develop high quality granular predictions, we develop a joint estimation framework dubbed OneMod that uses covariates to get an overall picture (e.g. extrapolation mode for locations with noisy/sparse data) and nonparametric estimates for high quality locations. OneMod was just used to obtain global all-cause mortality estimates published in the Lancet and we are currently scaling it to cause-specific mortality, a far more challenging task, order to support the Global Burden of Disease sstudy, one of the largest team science endeavors in the world. 

Dr. Aleksandr Aravkin received his PhD in Mathematics (Optimization) and master’s degree in statistics from the University of Washington in 2010. After some time at IBM TJ Watson Research center, he returned to the UW Applied Mathematics department in 2015, where he is currently an Associate Professor, and adjunct with Mathematics, Statistics, Computer Science, and Health Metrics Sciences. Since 2019, Dr. Aravkin has also held the title of Director of Mathematical Sciences at the Institute for Health Metrics and Evaluation (IHME), developing new methods and software implementations to solve emerging problems in global health.