Graph representations of solid state materials that encode only interatomic-distance information lack geometrical resolution,resulting in degenerate representations that may map distinct structures to equivalent graph...Graph representations of solid state materials that encode only interatomic-distance information lack geometrical resolution,resulting in degenerate representations that may map distinct structures to equivalent graphs.Here,we propose a hypergraph representation scheme for materials that allows for the association of higher-order geometrical information with hyperedges.Hyperedges generalize edges to connected sets of more than two nodes,and may be used to represent triplets and local environments of atoms in materials.This generalization of edges requires a different approach in graph convolution,which is developed in this work.These crystal hypergraph convolutional networks are trained based on various property prediction tasks for a vast set of solid-state materials available via MatBench.Results presented here focus on the improved performance of models based on both pairwise edges and local environment hyperedges.These results demonstrate that hypergraphs are an effective and efficient method for incorporating geometrical information in material representations.展开更多
Traffic prediction is a necessary function in intelligent transporta-tion systems to alleviate traffic congestion.Graph learning methods mainly focus on the spatiotemporal dimension,but ignore the nonlinear movement o...Traffic prediction is a necessary function in intelligent transporta-tion systems to alleviate traffic congestion.Graph learning methods mainly focus on the spatiotemporal dimension,but ignore the nonlinear movement of traffic prediction and the high-order relationships among various kinds of road segments.There exist two issues:1)deep integration of the spatiotempo-ral information and 2)global spatial dependencies for structural properties.To address these issues,we propose a nonlinear spatiotemporal optimization method,which introduces hypergraph convolution networks(HGCN).The method utilizes the higher-order spatial features of the road network captured by HGCN,and dynamically integrates them with the historical data to weigh the influence of spatiotemporal dependencies.On this basis,an extended Kalman filter is used to improve the accuracy of traffic prediction.In this study,a set of experiments were conducted on the real-world dataset in Chengdu,China.The result showed that the proposed method is feasible and accurate by two different time steps.Especially at the 15-minute time step,compared with the second-best method,the proposed method achieved 3.0%,11.7%,and 9.0%improvements in RMSE,MAE,and MAPE,respectively.展开更多
基金supported by the U.S.Department of Energy,Office of Science,Basic Energy Sciences,under Award No.DE-SC0023664This research used resources of the National Energy Research Scientific Computing Center(NERSC),a U.S.Department of Energy Office of Science User Facility located at Lawrence Berkeley National Laboratory,operated under Contract No.DE-AC02-05CH11231 using NERSC award BES-ERCAP0029544.
文摘Graph representations of solid state materials that encode only interatomic-distance information lack geometrical resolution,resulting in degenerate representations that may map distinct structures to equivalent graphs.Here,we propose a hypergraph representation scheme for materials that allows for the association of higher-order geometrical information with hyperedges.Hyperedges generalize edges to connected sets of more than two nodes,and may be used to represent triplets and local environments of atoms in materials.This generalization of edges requires a different approach in graph convolution,which is developed in this work.These crystal hypergraph convolutional networks are trained based on various property prediction tasks for a vast set of solid-state materials available via MatBench.Results presented here focus on the improved performance of models based on both pairwise edges and local environment hyperedges.These results demonstrate that hypergraphs are an effective and efficient method for incorporating geometrical information in material representations.
文摘Traffic prediction is a necessary function in intelligent transporta-tion systems to alleviate traffic congestion.Graph learning methods mainly focus on the spatiotemporal dimension,but ignore the nonlinear movement of traffic prediction and the high-order relationships among various kinds of road segments.There exist two issues:1)deep integration of the spatiotempo-ral information and 2)global spatial dependencies for structural properties.To address these issues,we propose a nonlinear spatiotemporal optimization method,which introduces hypergraph convolution networks(HGCN).The method utilizes the higher-order spatial features of the road network captured by HGCN,and dynamically integrates them with the historical data to weigh the influence of spatiotemporal dependencies.On this basis,an extended Kalman filter is used to improve the accuracy of traffic prediction.In this study,a set of experiments were conducted on the real-world dataset in Chengdu,China.The result showed that the proposed method is feasible and accurate by two different time steps.Especially at the 15-minute time step,compared with the second-best method,the proposed method achieved 3.0%,11.7%,and 9.0%improvements in RMSE,MAE,and MAPE,respectively.
