Session: Optimization in Global Health
Chair: Aleksandr Aravkin
Cluster: Optimization Applications (Communication, Energy, Health, ML, ...)
Talk 1: Large-scale Kernel Regression in Complex Global Health Estimation
Speaker: Peng Zheng
Abstract: We present an efficient approach to large-scale kernel regression with applications to estimating global mortality and causes of death. We highlight computational elements, particularly how computational elements in kernel regression interact with the problem scale and technical requirements, including constraints and aggregated observations.
Talk 2: Joint estimation of Prevalence, Sensitivity, and Specificity
Speaker: Nora Gilbertson
Abstract: Lack of perfect tests is a classic problem in epidemiology, and must be overcome to understand prevalence and burden of disease. Multiple imperfect tests are typically available, with partial information on their diagnostic properties (such as sensitivity and specificity). We present a joint inversion approach that allows us to obtain improved results for location-specific prevalence and diagnostic properties of multiple tests jointly, using all available information multiple locations and multiple imperfect tests. We explain the approach and show results on both simulated cases and schistosomiasis data.
Talk 3: Fast optimization approaches for raking
Speaker: Ariane Ducellier
Abstract: Raking is a classic problem in survey science, where available granular estimates are updated so that their aggregations across particular dimensions match available constraints from independent sources. We formulate raking as an optimization problem, and show how to efficiently solve complex raking examples in multiple dimensions with both direct and aggregated observations. The approach leverages duality theory, and intermediate results together with the implicit function theorem allow us to efficiently estimate asymptotic uncertainty of the raked estimates.
