CAMBRIDGE (MA) August, 2017. Minrui Zheng, a Ph.D. Student and researcher at UNCC’s Center for Applied GIScience, received an Honorable Mention in the Best Paper Competition at the 2nd International Symposium of Spatiotemporal Computing, held at Harvard University. Zheng, who focuses on advanced modeling approaches to understand explore complex problems, won the recognition with her paper, “Hyperparameter optimization of neural network-driven spatial models accelerated using cyber-enabled high-performance computing.”
With co-authors Drs. Wenwu Tang and Visiting Scholar Xiang Zhao, Zheng’s paper describes a method she developed to automate the parameterization of artificial neural network models (ANN), thereby avoiding a costly trial-and-error to design the network architectures of these increasingly important machine-learning algorithms.
Zheng and her collaborators wanted to look at how driving factors influenced land prices in Mecklenburg County (NC), one of the Nation’s largest and fastest growing regions. The problem was not trivial: estimating land prices required overcoming computational barriers associated with handling the sheer volume of geographic “big data”. Zheng used custom code and parallel computing, an acceleration method deployed on UNCC’s high-performance computing cluster COBRA, to automatically evaluate the wide range of options and optimize the design of an ANN to estimate price at specific locations and times.
Zheng’s approach reduced the processing time from 17 days to under 5 hours, substantially improving computer performance and greatly reducing a key bottleneck to using ANNs. In the future, Zheng hopes to improve the automation’s ability to deal with more and increasingly complex datum with similar savings in time and effort.