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
semantics information while maintaining spatial detail con-texts.Long-range context information plays a crucial role in this scenario.How-ever,the traditional convolution kernel only provides the local and small size ...semantics information while maintaining spatial detail con-texts.Long-range context information plays a crucial role in this scenario.How-ever,the traditional convolution kernel only provides the local and small size of the receptivefield.To address the problem,we propose a plug-and-play module aggregating both local and global information(aka LGIA module)to capture the high-order relationship between nodes that are far apart.We incorporate both local and global correlations into hypergraph which is able to capture high-order rela-tionships between nodes via the concept of a hyperedge connecting a subset of nodes.The local correlation considers neighborhood nodes that are spatially adja-cent and similar in the same CNN feature maps of magnetic resonance(MR)image;and the global correlation is searched from a batch of CNN feature maps of MR images in feature space.The influence of these two correlations on seman-tic segmentation is complementary.We validated our LGIA module on various CNN segmentation models with the cardiac MR images dataset.Experimental results demonstrate that our approach outperformed several baseline models.展开更多
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
The interactions between drugs and microbes affecting microbial abundance can lead to various diseases or reduce the effectiveness of pharmaceutical treatments.Traditional Microbe-Drug Association(MDA)determination th...The interactions between drugs and microbes affecting microbial abundance can lead to various diseases or reduce the effectiveness of pharmaceutical treatments.Traditional Microbe-Drug Association(MDA)determination through biological assays is time-consuming and costly.With the accumulation of MDA data,computational methods have become a promising approach to infer potential MDAs.Although existing methods focus on predicting whether a drug interacts with a microbe,they can rarely infer whether a drug promotes or inhibits the abundance of a given microbe.Moreover,the extreme imbalance among abundance-promoted,abundance-inhibited,and non-impacted cases remains a challenge for computational prediction methods.To address these issues,we propose a framework for predicting the imbalanced Impact of Drugs on Microbial Abundance by leveraging Multi-view Learning and Data Augmentation,named IDMA-MLDA.IDMA-MLDA employs a novel method of transforming a bipartite graph into a hypergraph,uses hypergraph convolutions to capture high-order vertex neighborhoods(macro-view),and employs graph neural networks to learn individual features of drugs and microbes(micro-view).It integrates features from both macro-view and micro-view to obtain more comprehensive representations,incorporates a data augmentation module to handle class imbalance,and uses a multilayer perceptron to predict the impact of drugs on microbial abundance.We demonstrate the superiority of IDMA-MLDA through comparisons with six baseline methods,and ablation studies affirm the contributions of each key module in IDMA-MLDA’s prediction.Furthermore,a comprehensive literature review verifies the abundance types of twelve MDAs predicted by IDMA-MLDA.展开更多
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.
基金supported by the Sichuan Science and Technology Program(Grant No.2019ZDZX0005,2019YFG0496,2020YFG0143,2019JDJQ0002 and 2020YFG0009).
文摘semantics information while maintaining spatial detail con-texts.Long-range context information plays a crucial role in this scenario.How-ever,the traditional convolution kernel only provides the local and small size of the receptivefield.To address the problem,we propose a plug-and-play module aggregating both local and global information(aka LGIA module)to capture the high-order relationship between nodes that are far apart.We incorporate both local and global correlations into hypergraph which is able to capture high-order rela-tionships between nodes via the concept of a hyperedge connecting a subset of nodes.The local correlation considers neighborhood nodes that are spatially adja-cent and similar in the same CNN feature maps of magnetic resonance(MR)image;and the global correlation is searched from a batch of CNN feature maps of MR images in feature space.The influence of these two correlations on seman-tic segmentation is complementary.We validated our LGIA module on various CNN segmentation models with the cardiac MR images dataset.Experimental results demonstrate that our approach outperformed several baseline models.
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.
基金supported by the National Natural Science Foundation of China(No.62372375)the Shaanxi Province Key R&D Program(No.2023-YBSF-114)the CAAI-Huawei MindSpore Open Fund(No.CAAIXSJLJJ-2022-035A).
文摘The interactions between drugs and microbes affecting microbial abundance can lead to various diseases or reduce the effectiveness of pharmaceutical treatments.Traditional Microbe-Drug Association(MDA)determination through biological assays is time-consuming and costly.With the accumulation of MDA data,computational methods have become a promising approach to infer potential MDAs.Although existing methods focus on predicting whether a drug interacts with a microbe,they can rarely infer whether a drug promotes or inhibits the abundance of a given microbe.Moreover,the extreme imbalance among abundance-promoted,abundance-inhibited,and non-impacted cases remains a challenge for computational prediction methods.To address these issues,we propose a framework for predicting the imbalanced Impact of Drugs on Microbial Abundance by leveraging Multi-view Learning and Data Augmentation,named IDMA-MLDA.IDMA-MLDA employs a novel method of transforming a bipartite graph into a hypergraph,uses hypergraph convolutions to capture high-order vertex neighborhoods(macro-view),and employs graph neural networks to learn individual features of drugs and microbes(micro-view).It integrates features from both macro-view and micro-view to obtain more comprehensive representations,incorporates a data augmentation module to handle class imbalance,and uses a multilayer perceptron to predict the impact of drugs on microbial abundance.We demonstrate the superiority of IDMA-MLDA through comparisons with six baseline methods,and ablation studies affirm the contributions of each key module in IDMA-MLDA’s prediction.Furthermore,a comprehensive literature review verifies the abundance types of twelve MDAs predicted by IDMA-MLDA.
