《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (8): 2409-2420.DOI: 10.11772/j.issn.1001-9081.2024081140
• 2024年全国开放式分布与并行计算学术年会 (DPCS 2024) • 上一篇
收稿日期:
2024-08-14
修回日期:
2024-09-14
接受日期:
2024-09-23
发布日期:
2024-09-25
出版日期:
2025-08-10
通讯作者:
任怡
作者简介:
涂银川(1996—),男,湖北武汉人,硕士研究生,主要研究方向:图神经网络、云计算、分布式机器学习基金资助:
Yinchuan TU, Yong GUO, Heng MAO, Yi REN(), Jianfeng ZHANG, Bao LI
Received:
2024-08-14
Revised:
2024-09-14
Accepted:
2024-09-23
Online:
2024-09-25
Published:
2025-08-10
Contact:
Yi REN
About author:
TU Yinchuan, born in 1996, M. S. candidate. His research interests include graph neural network, cloud computing, distributed machine learning.Supported by:
摘要:
随着图数据规模的快速增长,图神经网络(GNN)在处理大规模图结构数据时面临计算和存储方面的挑战。传统的单机训练方法已不足以应对日益庞大的数据集和复杂的GNN模型,分布式训练凭借并行计算能力和可扩展性,成为解决这些问题的有效途径。然而,一方面,已有的分布式GNN训练评估主要关注以模型精度为代表的性能指标和以训练时间为代表的效率指标,而较少关注数据处理效率和计算资源利用方面的指标;另一方面,算法效率评估的主要场景为单机单卡或单机多卡,而已有的评估方法在分布式环境中的应用相对简单。针对这些不足,提出针对分布式场景的模型训练的评估方法,涵盖评估指标、数据集和模型这3个方面。根据评估方法,选取3个代表性GNN模型,在4个具有不同数据特征的大型公开图数据集上进行分布式训练实验,并收集和分析得到的评估指标。实验结果表明,分布式训练中的模型架构和数据结构特征对模型复杂度、训练时间、计算节点吞吐量和计算节点平均吞吐量之比(NATR)均有影响;样本处理与数据拷贝占用了模型训练较多时间,计算节点互相等待的时间也不容忽视;相较于单机训练,分布式训练的计算节点吞吐量有显著降低,且需要进一步优化分布式系统中的资源利用。可见,所提评估方法为GNN模型在分布式环境中的训练性能优化提供了参考依据,并为模型的进一步优化和算法的改进奠定了实验基础。
中图分类号:
涂银川, 郭勇, 毛恒, 任怡, 张建锋, 李宝. 基于分布式环境的图神经网络模型训练效率与训练性能评估[J]. 计算机应用, 2025, 45(8): 2409-2420.
Yinchuan TU, Yong GUO, Heng MAO, Yi REN, Jianfeng ZHANG, Bao LI. Evaluation of training efficiency and training performance of graph neural network models based on distributed environment[J]. Journal of Computer Applications, 2025, 45(8): 2409-2420.
数据集 | 节点数 | 边数 | 平均度数 | 节点特征维度 | 类别数 | 应用领域 |
---|---|---|---|---|---|---|
CoraFull[ | 19 793 | 126 842 | 6.41 | 8 710 | 70 | 引文网络 |
ogbn-arxiv[ | 169 343 | 1 166 243 | 6.89 | 128 | 40 | 引文网络 |
Reddit[ | 232 965 | 114 615 892 | 491.99 | 602 | 41 | 社交网络 |
ogbn-products[ | 2 449 029 | 61 859 140 | 25.26 | 100 | 47 | 推荐系统 |
表1 评估中选择的数据集
Tab. 1 Datasets selected in evaluation
数据集 | 节点数 | 边数 | 平均度数 | 节点特征维度 | 类别数 | 应用领域 |
---|---|---|---|---|---|---|
CoraFull[ | 19 793 | 126 842 | 6.41 | 8 710 | 70 | 引文网络 |
ogbn-arxiv[ | 169 343 | 1 166 243 | 6.89 | 128 | 40 | 引文网络 |
Reddit[ | 232 965 | 114 615 892 | 491.99 | 602 | 41 | 社交网络 |
ogbn-products[ | 2 449 029 | 61 859 140 | 25.26 | 100 | 47 | 推荐系统 |
模型 | 基本概念 | 参数共享 | 计算效率 | 灵活性 | 抗过平滑 |
---|---|---|---|---|---|
GCN | 图卷积 | 是 | 高 | 低 | 低 |
GAT | 注意力 | 否 | 中 | 高 | 高 |
GraphSAGE | 采样聚合 | 是 | 高 | 高 | 中 |
表2 评估中选择的模型
Tab. 2 Models selected in evaluation
模型 | 基本概念 | 参数共享 | 计算效率 | 灵活性 | 抗过平滑 |
---|---|---|---|---|---|
GCN | 图卷积 | 是 | 高 | 低 | 低 |
GAT | 注意力 | 否 | 中 | 高 | 高 |
GraphSAGE | 采样聚合 | 是 | 高 | 高 | 中 |
模型 | 聚合方式 | 学习率 | 激活函数 | dropout |
---|---|---|---|---|
GCN | 求和 | 0.003 | ReLU | 0.5 |
GAT | 注意力机制 | 0.003 | ReLU | 0.5 |
GraphSAGE | 均值 | 0.003 | ReLU | 0.5 |
表3 模型的参数设置
Tab. 3 Model parameter setting
模型 | 聚合方式 | 学习率 | 激活函数 | dropout |
---|---|---|---|---|
GCN | 求和 | 0.003 | ReLU | 0.5 |
GAT | 注意力机制 | 0.003 | ReLU | 0.5 |
GraphSAGE | 均值 | 0.003 | ReLU | 0.5 |
数据集 | 隐藏层维度 | 数据集 | 隐藏层维度 |
---|---|---|---|
CoraFull | 64 | 512 | |
ogbn-arxiv | 256 | ogbn-products | 1 024 |
表4 数据集对应的隐藏层维度
Tab. 4 Hidden layer dimensions for datasets
数据集 | 隐藏层维度 | 数据集 | 隐藏层维度 |
---|---|---|---|
CoraFull | 64 | 512 | |
ogbn-arxiv | 256 | ogbn-products | 1 024 |
数据集名称 | 分区数量 | 分区特征 | 分区1 | 分区2 | 分区3 | 分区4 | 分区5 | 分区6 | 分区7 | 分区8 |
---|---|---|---|---|---|---|---|---|---|---|
CoraFull | 4 | 节点 | 6 245 | 7 054 | 6 580 | 6 056 | — | — | — | — |
5 095 | 5 095 | 5 095 | 4 584 | |||||||
边 | 33 962 | 35 316 | 33 994 | 35 482 | — | — | — | — | ||
31 797 | 31 656 | 30 933 | 32 456 | |||||||
8 | 节点 | 3 548 | 3 979 | 3 896 | 3 561 | 3 776 | 3 576 | 4 031 | 3 681 | |
2 545 | 2 546 | 2 396 | 2 161 | 2 530 | 2 539 | 2 543 | 2 533 | |||
边 | 17 482 | 18 838 | 19 334 | 17 852 | 17 304 | 18 008 | 18 454 | 18 182 | ||
15 783 | 16 305 | 16 216 | 14 874 | 15 312 | 16 184 | 16 040 | 16 128 | |||
ogbn-arxiv | 4 | 节点 | 72 166 | 73 705 | 79 222 | 69 758 | — | — | — | — |
42 151 | 43 591 | 42 842 | 40 759 | |||||||
边 | 349 845 | 386 028 | 408 835 | 410 273 | — | — | — | — | ||
304 982 | 343 654 | 343 477 | 343 473 | |||||||
8 | 节点 | 50 225 | 49 622 | 45 405 | 44 941 | 45 008 | 50 709 | 49 064 | 45 018 | |
21 777 | 21 801 | 19 599 | 21 449 | 21 776 | 20 912 | 20 229 | 21 800 | |||
边 | 215 378 | 212 008 | 223 712 | 213 756 | 186 270 | 208 424 | 201 564 | 192 607 | ||
171 879 | 171 701 | 171 928 | 171 784 | 153 345 | 171 954 | 159 784 | 163 211 | |||
4 | 节点 | 203 689 | 167 999 | 175 006 | 168 965 | — | — | — | — | |
53 928 | 59 455 | 59 570 | 60 012 | |||||||
边 | 34 882 726 | 33 785 415 | 33 495 268 | 33 354 912 | — | — | — | — | ||
29 214 769 | 28 417 939 | 28 841 090 | 28 375 059 | |||||||
8 | 节点 | 140 091 | 124 022 | 178 234 | 141 059 | 121 485 | 166 820 | 117 723 | 150 886 | |
30 030 | 25 490 | 30 035 | 29 887 | 29 800 | 29 914 | 29 416 | 28 393 | |||
边 | 17 578 366 | 19 521 642 | 20 797 883 | 19 818 501 | 16 708 704 | 16 124 892 | 16 446 438 | 19 275 925 | ||
13 865 509 | 14 724 005 | 14 803 007 | 14 736 263 | 14 762 642 | 12 743 840 | 14 412 564 | 14 801 027 | |||
ogbn-products | 4 | 节点 | 1 073 131 | 992 338 | 1 012 637 | 972 605 | — | — | — | — |
610 657 | 615 415 | 592 322 | 630 635 | |||||||
边 | 35 653 651 | 34 895 467 | 31 336 778 | 34 031 337 | — | — | — | — | ||
32 481 110 | 32 374 997 | 29 612 148 | 31 698 798 | |||||||
8 | 节点 | 566 183 | 553 209 | 578 443 | 61 531 | 526 883 | 606 943 | 628 995 | 609 695 | |
315 327 | 309 400 | 310 846 | 261 228 | 315 322 | 314 885 | 307 820 | 314 201 | |||
边 | 16 941 573 | 16 402 542 | 17 432 828 | 17 087 742 | 17 007 390 | 17 469 477 | 17 848 086 | 17 167 751 | ||
15 681 661 | 15 388 305 | 15 815 285 | 15 341 841 | 16 084 289 | 15 962 365 | 16 113 958 | 15 779 349 |
表5 数据划分结果
Tab. 5 Results of data partition
数据集名称 | 分区数量 | 分区特征 | 分区1 | 分区2 | 分区3 | 分区4 | 分区5 | 分区6 | 分区7 | 分区8 |
---|---|---|---|---|---|---|---|---|---|---|
CoraFull | 4 | 节点 | 6 245 | 7 054 | 6 580 | 6 056 | — | — | — | — |
5 095 | 5 095 | 5 095 | 4 584 | |||||||
边 | 33 962 | 35 316 | 33 994 | 35 482 | — | — | — | — | ||
31 797 | 31 656 | 30 933 | 32 456 | |||||||
8 | 节点 | 3 548 | 3 979 | 3 896 | 3 561 | 3 776 | 3 576 | 4 031 | 3 681 | |
2 545 | 2 546 | 2 396 | 2 161 | 2 530 | 2 539 | 2 543 | 2 533 | |||
边 | 17 482 | 18 838 | 19 334 | 17 852 | 17 304 | 18 008 | 18 454 | 18 182 | ||
15 783 | 16 305 | 16 216 | 14 874 | 15 312 | 16 184 | 16 040 | 16 128 | |||
ogbn-arxiv | 4 | 节点 | 72 166 | 73 705 | 79 222 | 69 758 | — | — | — | — |
42 151 | 43 591 | 42 842 | 40 759 | |||||||
边 | 349 845 | 386 028 | 408 835 | 410 273 | — | — | — | — | ||
304 982 | 343 654 | 343 477 | 343 473 | |||||||
8 | 节点 | 50 225 | 49 622 | 45 405 | 44 941 | 45 008 | 50 709 | 49 064 | 45 018 | |
21 777 | 21 801 | 19 599 | 21 449 | 21 776 | 20 912 | 20 229 | 21 800 | |||
边 | 215 378 | 212 008 | 223 712 | 213 756 | 186 270 | 208 424 | 201 564 | 192 607 | ||
171 879 | 171 701 | 171 928 | 171 784 | 153 345 | 171 954 | 159 784 | 163 211 | |||
4 | 节点 | 203 689 | 167 999 | 175 006 | 168 965 | — | — | — | — | |
53 928 | 59 455 | 59 570 | 60 012 | |||||||
边 | 34 882 726 | 33 785 415 | 33 495 268 | 33 354 912 | — | — | — | — | ||
29 214 769 | 28 417 939 | 28 841 090 | 28 375 059 | |||||||
8 | 节点 | 140 091 | 124 022 | 178 234 | 141 059 | 121 485 | 166 820 | 117 723 | 150 886 | |
30 030 | 25 490 | 30 035 | 29 887 | 29 800 | 29 914 | 29 416 | 28 393 | |||
边 | 17 578 366 | 19 521 642 | 20 797 883 | 19 818 501 | 16 708 704 | 16 124 892 | 16 446 438 | 19 275 925 | ||
13 865 509 | 14 724 005 | 14 803 007 | 14 736 263 | 14 762 642 | 12 743 840 | 14 412 564 | 14 801 027 | |||
ogbn-products | 4 | 节点 | 1 073 131 | 992 338 | 1 012 637 | 972 605 | — | — | — | — |
610 657 | 615 415 | 592 322 | 630 635 | |||||||
边 | 35 653 651 | 34 895 467 | 31 336 778 | 34 031 337 | — | — | — | — | ||
32 481 110 | 32 374 997 | 29 612 148 | 31 698 798 | |||||||
8 | 节点 | 566 183 | 553 209 | 578 443 | 61 531 | 526 883 | 606 943 | 628 995 | 609 695 | |
315 327 | 309 400 | 310 846 | 261 228 | 315 322 | 314 885 | 307 820 | 314 201 | |||
边 | 16 941 573 | 16 402 542 | 17 432 828 | 17 087 742 | 17 007 390 | 17 469 477 | 17 848 086 | 17 167 751 | ||
15 681 661 | 15 388 305 | 15 815 285 | 15 341 841 | 16 084 289 | 15 962 365 | 16 113 958 | 15 779 349 |
数据集 | 模型 | 总时间/s | 样本采样与数据拷贝 | 前向传播 | 反向传播 | 参数更新 | ||||
---|---|---|---|---|---|---|---|---|---|---|
时间/s | 占比/% | 时间/s | 占比/% | 时间/s | 占比/% | 时间/s | 占比/% | |||
CoraFull | GCN | 26.84 | 5.98 | 22.28 | 4.69 | 17.47 | 1.50 | 5.59 | 1.03 | 3.84 |
GAT | 125.47 | 19.12 | 15.24 | 18.81 | 14.99 | 10.35 | 8.25 | 2.93 | 2.34 | |
GraphSAGE | 178.64 | 23.29 | 13.04 | 19.11 | 10.70 | 10.23 | 5.73 | 2.06 | 1.15 | |
ogbn-arxiv | GCN | 44.98 | 12.18 | 27.08 | 8.68 | 19.30 | 2.62 | 5.82 | 1.41 | 3.13 |
GAT | 186.95 | 40.35 | 21.58 | 33.54 | 17.94 | 16.59 | 8.87 | 3.06 | 1.64 | |
GraphSAGE | 256.18 | 63.97 | 24.97 | 38.65 | 15.09 | 15.83 | 6.18 | 2.48 | 0.97 | |
GCN | 372.26 | 34.47 | 9.26 | 20.20 | 5.43 | 5.98 | 1.61 | 1.43 | 0.38 | |
GAT | 569.16 | 35.93 | 6.31 | 31.45 | 5.53 | 13.02 | 2.29 | 2.51 | 0.44 | |
GraphSAGE | 811.44 | 45.84 | 5.65 | 27.05 | 3.33 | 11.56 | 1.42 | 2.34 | 0.29 | |
ogbn-products | GCN | 295.20 | 58.39 | 19.78 | 22.82 | 7.73 | 7.34 | 2.49 | 1.69 | 0.57 |
GAT | 438.68 | 65.00 | 14.82 | 52.34 | 11.93 | 23.63 | 5.39 | 3.68 | 0.84 | |
GraphSAGE | 599.61 | 76.82 | 12.81 | 54.42 | 9.08 | 24.84 | 4.14 | 3.27 | 0.55 |
表6 在4个计算节点上模型训练10个周期的时间消耗及时间分解
Tab. 6 Time consumption and decomposition for 10 epochs in model training with 4 computing nodes
数据集 | 模型 | 总时间/s | 样本采样与数据拷贝 | 前向传播 | 反向传播 | 参数更新 | ||||
---|---|---|---|---|---|---|---|---|---|---|
时间/s | 占比/% | 时间/s | 占比/% | 时间/s | 占比/% | 时间/s | 占比/% | |||
CoraFull | GCN | 26.84 | 5.98 | 22.28 | 4.69 | 17.47 | 1.50 | 5.59 | 1.03 | 3.84 |
GAT | 125.47 | 19.12 | 15.24 | 18.81 | 14.99 | 10.35 | 8.25 | 2.93 | 2.34 | |
GraphSAGE | 178.64 | 23.29 | 13.04 | 19.11 | 10.70 | 10.23 | 5.73 | 2.06 | 1.15 | |
ogbn-arxiv | GCN | 44.98 | 12.18 | 27.08 | 8.68 | 19.30 | 2.62 | 5.82 | 1.41 | 3.13 |
GAT | 186.95 | 40.35 | 21.58 | 33.54 | 17.94 | 16.59 | 8.87 | 3.06 | 1.64 | |
GraphSAGE | 256.18 | 63.97 | 24.97 | 38.65 | 15.09 | 15.83 | 6.18 | 2.48 | 0.97 | |
GCN | 372.26 | 34.47 | 9.26 | 20.20 | 5.43 | 5.98 | 1.61 | 1.43 | 0.38 | |
GAT | 569.16 | 35.93 | 6.31 | 31.45 | 5.53 | 13.02 | 2.29 | 2.51 | 0.44 | |
GraphSAGE | 811.44 | 45.84 | 5.65 | 27.05 | 3.33 | 11.56 | 1.42 | 2.34 | 0.29 | |
ogbn-products | GCN | 295.20 | 58.39 | 19.78 | 22.82 | 7.73 | 7.34 | 2.49 | 1.69 | 0.57 |
GAT | 438.68 | 65.00 | 14.82 | 52.34 | 11.93 | 23.63 | 5.39 | 3.68 | 0.84 | |
GraphSAGE | 599.61 | 76.82 | 12.81 | 54.42 | 9.08 | 24.84 | 4.14 | 3.27 | 0.55 |
模型 | CoraFull | ogbn-arxiv | ogbn-products | |
---|---|---|---|---|
GCN | 38.83 | 8.19 | 15.69 | 14.26 |
GAT | 9.82 | 3.01 | 8.39 | 5.67 |
GraphSAGE | 5.92 | 1.50 | 5.46 | 4.87 |
表7 用4个计算节点训练的NATR (%)
Tab. 7 NATR trained with 4 computing nodes
模型 | CoraFull | ogbn-arxiv | ogbn-products | |
---|---|---|---|---|
GCN | 38.83 | 8.19 | 15.69 | 14.26 |
GAT | 9.82 | 3.01 | 8.39 | 5.67 |
GraphSAGE | 5.92 | 1.50 | 5.46 | 4.87 |
[1] | YING R, HE R, CHEN K, et al. Graph convolutional neural networks for web-scale recommender systems[C]// Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2018: 974-983. |
[2] | VATTER J, MAYER R, JACOBSEN H A. The evolution of distributed systems for graph neural networks and their origin in graph processing and deep learning: a survey[J]. ACM Computing Surveys, 2024, 56(1): No.6. |
[3] | SHAO Y, LI H, GU X, et al. Distributed graph neural network training: a survey[J]. ACM Computing Surveys, 2024, 56(8): No.191. |
[4] | DWIVEDI V P, JOSHI C K, LUU A T, et al. Benchmarking graph neural networks[J]. Journal of Machine Learning Research, 2023, 23: 1-48. |
[5] | 魏嘉,张兴军,王龙翔,等. 面向深度神经网络大规模分布式数据并行训练的MC2能耗模型[J]. 计算机研究与发展, 2024, 61(12): 2985-3004. |
WEI J, ZHANG X J, WANG L X, et al. MC2 energy consumption model for massively distributed data parallel training of deep neural network[J]. Journal of Computer Research and Development, 2024, 61(12): 2985-3004. | |
[6] | WU J, SUN J, SUN H, et al. Performance analysis of graph neural network frameworks[C]// Proceedings of the 2021 IEEE International Symposium on Performance Analysis of Systems and Software. Piscataway: IEEE, 2021: 118-127. |
[7] | ZHANG L, LU K, LAI Z, et al. Accelerating GNN training by adapting large graphs to distributed heterogeneous architectures[J]. IEEE Transactions on Computers, 2023, 72(12): 3473-3488. |
[8] | ZHANG L, LAI Z, TANG Y, et al. PCGraph: accelerating GNN inference on large graphs via partition caching[C]// Proceedings of the 19th IEEE International Symposium on Parallel and Distributed Processing with Applications/ 11th IEEE International Conference on Big Data and Cloud Computing/ 14th IEEE International Conference on Social Computing and Networking/ 11th IEEE International Conference on Sustainable Computing and Communications. Piscataway: IEEE, 2021: 279-287. |
[9] | LIN H, YAN M, YE X, et al. A comprehensive survey on distributed training of graph neural networks[J]. Proceedings of the IEEE, 2023, 111(12): 1572-1606. |
[10] | MA L, YANG Z, MIAO Y, et al. NeuGraph: parallel deep neural network computation on large graphs[C]// Proceedings of the 2019 USENIX Annual Technical Conference. Berkeley: USENIX Association, 2019: 443-457. |
[11] | CAI Z, ZHOU Q, YAN X, et al. DSP: efficient GNN training with multiple GPUs[C]// Proceedings of the 28th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming. New York: ACM, 2023: 392-404. |
[12] | WU X, SHI L, HE L, et al. TurboGNN: improving the end-to-end performance for sampling-based GNN training on GPUs[J]. IEEE Transactions on Computers, 2023, 72(9): 2571-2584. |
[13] | WU W, SHI X, HE L, et al. TurboMGNN: improving concurrent GNN training tasks on GPU with fine-grained kernel fusion[J]. IEEE Transactions on Parallel and Distributed Systems, 2023, 34(6): 1968-1981. |
[14] | SUN J, SU L, SHI Z, et al. Legion: automatically pushing the envelope of multi-GPU system for billion-scale GNN training[C]// Proceedings of the 2023 USENIX Annual Technical Conference. Berkeley: USENIX Association, 2023: 165-179. |
[15] | WAN B, ZHAO J, WU C. Adaptive message quantization and parallelization for distributed full-graph GNN training[EB/OL]. [2024-07-15].. |
[16] | LIU H, LU S, CHEN X, et al. G3: when Graph neural networks meet parallel Graph processing systems on GPUs[J]. Proceedings of the VLDB Endowment, 2020, 13(12): 2813-2816. |
[17] | KALER T, ILIOPOULOS A S, MURZYNOWSKI P, et al. Communication-efficient graph neural networks with probabilistic neighborhood expansion analysis and caching[EB/OL]. [2024-07-15].. |
[18] | TEKDOĞAN T, GOKTAŞ S, YILMAZER-METIN A. gSuite: a flexible and framework independent benchmark suite for graph neural network inference on GPUs[C]// Proceedings of the 2022 IEEE International Symposium on Workload Characterization. Piscataway: IEEE, 2022: 146-159. |
[19] | CHEN T, ZHOU K, DUAN K, et al. Bag of tricks for training deeper graph neural networks: a comprehensive benchmark study[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(3): 2769-2781. |
[20] | 赵港,王千阁,姚烽,等. 大规模图神经网络系统综述[J]. 软件学报, 2022, 33(1): 150-170. |
ZHAO G, WANG Q G, YAO F, et al. Survey on large-scale graph neural network systems[J]. Journal of Software, 2022, 33(1): 150-170. | |
[21] | BARUAH T, SHIVDIKAR K, DONG S, et al. GNNMark: a benchmark suite to characterize graph neural network training on GPUs[C]// Proceedings of the 2021 IEEE International Symposium on Performance Analysis of Systems and Software. Piscataway: IEEE, 2021: 13-23. |
[22] | LV Q, DING M, LIU Q, et al. Are we really making much progress? revisiting, benchmarking and refining heterogeneous graph neural networks[C]// Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York: ACM, 2021: 1150-1160. |
[23] | WU F, DE SOUZA A H, Jr, ZHANG T, et al. Simplifying graph convolutional networks[C]// Proceedings of the 36th International Conference on Machine Learning. New York: JMLR.org, 2019: 6861-6871. |
[24] | KIPF T N, WELLING M. Semi-supervised classification with graph convolutional networks[EB/OL]. [2024-07-15].. |
[25] | GILMER J, SCHOENHOLZ S S, RILEY P F, et al. Neural message passing for quantum chemistry[C]// Proceedings of the 34th International Conference on Machine Learning. New York: JMLR.org, 2017: 1263-1272. |
[26] | DEARING M T, WANG X. Analyzing the performance of graph neural networks with pipe parallelism[EB/OL]. [2024-07-15].. |
[27] | LIN H, YAN M, YANG X, et al. Characterizing and understanding distributed GNN training on GPUs[J]. IEEE Computer Architecture Letters, 2022, 21(1): 21-24. |
[28] | WANG Z, WANG Y, YUAN C, et al. Empirical analysis of performance bottlenecks in graph neural network training and inference with GPUs[J]. Neurocomputing, 2021, 446: 165-191. |
[29] | SHCHUR O, MUMME M, BOJCHEVSKI A, et al. Pitfalls of graph neural network evaluation[EB/OL]. [2024-07-15].. |
[30] | HU W, FEY M, ZITNIK M, et al. Open graph benchmark: datasets for machine learning on graphs[C]// Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2020: 22118-22133. |
[31] | NVIDIA Developer. NVIDIA visual profiler[EB/OL]. [2024-07-15].. |
[32] | VILLA O, STEPHENSON M, NELLANS D, et al. NVBit: a dynamic binary instrumentation framework for NVIDIA GPUs[C]// Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture. New York: ACM, 2019: 372-383. |
[33] | ZHENG D, MA C, WANG M, et al. DistDGL: distributed graph neural network training for billion-scale graphs[C]// Proceedings of the IEEE/ACM 10th Workshop on Irregular Applications: Architectures and Algorithms. Piscataway: IEEE, 2020: 36-44. |
[34] | KETKAR R, LIU Y, WANG H, et al. A benchmark study of graph models for molecular acute toxicity prediction[J]. International Journal of Molecular Sciences, 2023, 24(15): No.11966. |
[35] | FEY M, LENSSEN J E. Fast graph representation learning with PyTorch Geometric[EB/OL]. [2024-07-15].. |
[36] | FUNG V, ZHANG J, JUAREZ E, et al. Benchmarking graph neural networks for materials chemistry[J]. npj Computational Materials, 2021, 7: No.84. |
[37] | LI F, FENG J, YAN H, et al. Dynamic graph convolutional recurrent network for traffic prediction: benchmark and solution[J]. ACM Transactions on Knowledge Discovery from Data, 2023, 17(1): No.9. |
[38] | CoraFullDataset — DGL 2.2.1 documentation[DS/OL]. [2024-07-15].. |
[39] | RedditDataset — DGL 2.2.1 documentation[DS/OL]. [2024-07-15].. |
[40] | VELIČKOVIĆ P, CUCURULL G, CASANOVA A, et al. Graph attention networks[EB/OL]. [2024-07-15].. |
[41] | HAMILTON W L, YING R, LESKOVEC J. Inductive representation learning on large graphs[C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 1025-1035. |
[42] | PASZKE A, GROSS S, MASSA F, et al. PyTorch: an imperative style, high-performance deep learning library[C]// Proceedings of the 33rd International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2019: 8026-8037. |
[43] | WANG M, ZHENG D, YE Z, et al. Deep graph library: a graph-centric, highly-performant package for graph neural networks[EB/OL]. [2024-07-15].. |
[44] | ABADI M, AGARWAL A, BARHAM P, et al. TensorFlow: large-scale machine learning on heterogeneous distributed systems[EB/OL]. [2024-07-15].. |
[45] | CHEN T, LI M, LI Y, et al. MXNet: a flexible and efficient machine learning library for heterogeneous distributed systems[EB/OL]. [2024-07-15].. |
[1] | 陈丹阳, 张长伦. 多尺度去相关的图卷积网络模型[J]. 《计算机应用》唯一官方网站, 2025, 45(7): 2180-2187. |
[2] | 张悦岚, 苏静, 赵航宇, 杨白利. 基于知识感知与交互的多视图蒸馏推荐算法[J]. 《计算机应用》唯一官方网站, 2025, 45(7): 2211-2220. |
[3] | 梁辰, 王奕森, 魏强, 杜江. 基于Tsransformer-GCN的源代码漏洞检测方法[J]. 《计算机应用》唯一官方网站, 2025, 45(7): 2296-2303. |
[4] | 张子墨, 赵雪专. 多尺度稀疏图引导的视觉图神经网络[J]. 《计算机应用》唯一官方网站, 2025, 45(7): 2188-2194. |
[5] | 姜超英, 李倩, 刘宁, 刘磊, 崔立真. 基于图对比学习的再入院预测模型[J]. 《计算机应用》唯一官方网站, 2025, 45(6): 1784-1792. |
[6] | 党伟超, 温鑫瑜, 高改梅, 刘春霞. 基于多视图多尺度对比学习的图协同过滤[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1061-1068. |
[7] | 王聪, 史艳翠. 基于多视角学习的图神经网络群组推荐模型[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1205-1212. |
[8] | 田仁杰, 景明利, 焦龙, 王飞. 基于混合负采样的图对比学习推荐算法[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1053-1060. |
[9] | 游兰, 张雨昂, 刘源, 陈智军, 王伟, 曾星, 何张玮. 基于协作贡献网络的开源项目开发者推荐[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1213-1222. |
[10] | 蔡启健, 谭伟. 语义图增强的多模态推荐算法[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 421-427. |
[11] | 马汉达, 吴亚东. 多域时空层次图神经网络的空气质量预测[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 444-452. |
[12] | 余肖生, 王智鑫. 基于多层次图对比学习的序列推荐模型[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 106-114. |
[13] | 程子栋, 李鹏, 朱枫. 物联网威胁情报知识图谱中潜在关系的挖掘[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 24-31. |
[14] | 赵文博, 马紫彤, 杨哲. 基于有向超图自适应卷积的链接预测模型[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 15-23. |
[15] | 唐廷杰, 黄佳进, 秦进. 基于图辅助学习的会话推荐[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2711-2718. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||