CAREER: Data-Centric Evolutionary Contagion Models with Parallel and Quantum Parallel Computing

点击翻译按钮获取中文摘要

Global viral epidemics produce vast amounts of high-dimensional data indexed by the location and time each virus is observed. Scientists, businesses, governments and independent organizations want to learn from this data so they can understand basic biological mechanisms, invest capital, allocate aid and design coherent policy in a changing world. Evolutionary contagion (Evo-Con) models seek to identify and predict viral variants with heightened rates of spread by jointly modeling viral contagion and evolution. Analyzing spatial patterns of viral contagion is an area of immense scientific interest, but the task requires accounting for the nature of transportation networks that shape the global economy. As a result, big data applications become computationally intensive and benefit from high-performance computing. The project advances knowledge and utility of Evo-Con models in the context of massive amounts of complex, dynamic and geographically distributed data. In lockstep with these developments, the investigator will develop "Statistical Learning Goes Viral", a free-access MOOC (Massive Open Online Course), with concomitant free-access expository textbook and further expand his efforts building bridges between UCLA and key historically black colleges and universities. The investigator will combine theory, methods and computing in a way that facilitates high-impact data analysis and easy measurement of success. The PI will (1) develop a class of nonlinear and multivariate phylogenetic Hawkes processes that use autoregressive neural networks to maintain both flexibility and scalability. Nonlinear phylogenetic Hawkes process development will capitalize on experience building heavily hierarchical models for joint data, but a pragmatic approach will depart from previous Bayesian implementations to enhance scalability and prediction; and (2) adapt these stochastic process models to complex transportation patterns in a way that responds to spatial data precision by combining nonlinear dimension reduction with convolutional neural networks (CNN). When travel networks explicitly present themselves, the PI will leverage experience building spatial models that account for network structures by incorporating nonlinearities through graph CNNs. More radical within phylogeography, the PI will eschew explicit network representations with the help of spherical CNNs that build implicit representations of global spatial dependencies to model viral contagion with increased flexibility. By hierarchically combining (1) and (2) within the same factor graph, the joint data neural model will fully integrate spatial, genomic, and temporal data. Finally, the investigator will (3) construct high-performance computing techniques that leverage conventional and quantum resources to fit complex, multimodal and high-dimensional model geometries. Computational developments will go well beyond track-record inventions of parallelized Markov chain Monte Carlo algorithms that use graphics processing units and quantum computers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.