ICML 2022spotlight49 citations

GraphFM: Improving Large-Scale GNN Training via Feature Momentum

Haiyang Yu, Limei Wang, Bokun Wang, Meng Liu, Tianbao Yang, Shuiwang Ji

Abstract

Training of graph neural networks (GNNs) for large-scale node classification is challenging. A key difficulty lies in obtaining accurate hidden node representations while avoiding the neighborhood explosion problem. Here, we propose a new technique, named feature momentum (FM), that uses a momentum step to incorporate historical embeddings when updating feature representations. We develop two specific algorithms, known as GraphFM-IB and GraphFM-OB, that consider in-batch and out-of-batch data, respectively. GraphFM-IB applies FM to in-batch sampled data, while GraphFM-OB applies FM to out-of-batch data that are 1-hop neighborhood of in-batch data. We provide a convergence analysis for GraphFM-IB and some theoretical insight for GraphFM-OB. Empirically, we observe that GraphFM-IB can effectively alleviate the neighborhood explosion problem of existing methods. In addition, GraphFM-OB achieves promising performance on multiple large-scale graph datasets.

BibTeX
@InProceedings{pmlr-v162-yu22g,
  title = 	 {{G}raph{FM}: Improving Large-Scale {GNN} Training via Feature Momentum},
  author =       {Yu, Haiyang and Wang, Limei and Wang, Bokun and Liu, Meng and Yang, Tianbao and Ji, Shuiwang},
  booktitle = 	 {Proceedings of the 39th International Conference on Machine Learning},
  pages = 	 {25684--25701},
  year = 	 {2022},
  editor = 	 {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan},
  volume = 	 {162},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {17--23 Jul},
  publisher =    {PMLR},
  pdf = 	 {https://proceedings.mlr.press/v162/yu22g/yu22g.pdf},
  url = 	 {https://proceedings.mlr.press/v162/yu22g.html},
  abstract = 	 {Training of graph neural networks (GNNs) for large-scale node classification is challenging. A key difficulty lies in obtaining accurate hidden node representations while avoiding the neighborhood explosion problem. Here, we propose a new technique, named feature momentum (FM), that uses a momentum step to incorporate historical embeddings when updating feature representations. We develop two specific algorithms, known as GraphFM-IB and GraphFM-OB, that consider in-batch and out-of-batch data, respectively. GraphFM-IB applies FM to in-batch sampled data, while GraphFM-OB applies FM to out-of-batch data that are 1-hop neighborhood of in-batch data. We provide a convergence analysis for GraphFM-IB and some theoretical insight for GraphFM-OB. Empirically, we observe that GraphFM-IB can effectively alleviate the neighborhood explosion problem of existing methods. In addition, GraphFM-OB achieves promising performance on multiple large-scale graph datasets.}
}
GraphFM: Improving Large-Scale GNN Training via Feature Momentum · ICML 2022