Four Duke Statistical Science graduate students were among the winners of the American Statistical Association’s Student Paper Competition. By winning the competition, these students will receive travel awards to attend the upcoming 2023 Joint Statistical Meetings (JSM) in Toronto, Ontario, Canada.
Bersson, a Ph.D. student advised by Professor Peter Hoff, won in the Social Statistics (SSS), Government Statistics (GSS), and Survey Research Methods (SRMS) section for her paper: “Optimal Conformal Prediction for Small Areas." Papers entered in this category must involve either a new statistical methodology or a creative application of statistical analyses to a problem, issue, or policy question pertinent to the subject areas in either of the three areas – SSS, GSS or SRMS.
Bersson’s article shows how, in demographic analyses, statistical predictions for a given area (such as a county) can make use of information from neighboring areas.
Jiang, a Master’s student studying with Professor Jason Xu, won in the Statistical Computing or Graphics (SCSG) section for his paper: “The Stochastic Proximal Distance Algorithm.” In this category, papers should involve some aspect of statistical computing, which might be original methodological research, a novel application, or a software-related project.
Jiang’s article uses techniques from stochastic approximation to present a version of the proximal distance algorithm, a method for solving statistical problems subject to general parameter constraints, that is suitable for large-scale, high-dimensional settings. He establishes convergence and finite-sample properties of the method that require novel theoretical arguments, as the baseline algorithm he extends relies on geometry that no longer holds in the stochastic setting.
Jin, a Ph.D. student advised by Professors David Dunson and Amy Herring, won in the Statistics and the Environment (ENVR) section for her paper, “Spatial Predictions on Physically Constrained Domains: Applications to Arctic Sea Salinity Data.” In this category, papers may consist of novel approaches to the analysis of environmental data, new methodology with a clear application to a statistical problem found within the environmental sciences, or an interesting application of statistics to environmental research.
Jin’s article develops a novel approach for modeling of spatiotemporal processes, which is designed to take into account the types of restricted domains that are common in many applications areas, while also facilitating efficient computation for large datasets. Jin applies her method to predict sea surface salinity (SSS) in the Arctic Ocean, a crucial indicator of climate change.
Zhu, a Ph.D. student advised by Professor David Dunson, won in Bayesian Statistical Science (SBSS) for his paper: Radial Neighbors for Provably Accurate Scalable Approximations of Gaussian Processes. This category – which is for completed research – highlights research on Bayesian methodology, which may be broadly construed and includes applied, computational, or theoretical work.
Gaussian processes provide a very important tool for modeling of spatiotemporal processes and unknown input-output relationships in broad application areas ranging from environmental sciences to physics. To overcome computational bottlenecks arising in massive data settings, it has become common to rely on sparse approximations characterized via a graph. However, such approaches may be sensitive to graph choice and lack theoretical guarantees. Zhu’s article provides a major breakthrough by developing rigorous approximation guarantees for a novel choice of graph.
Statistical Science
