Timezone: »

Local Augmentation for Graph Neural Networks
Songtao Liu · Rex (Zhitao) Ying · Hanze Dong · Lanqing Li · Tingyang Xu · Yu Rong · Peilin Zhao · Junzhou Huang · Dinghao Wu

Wed Jul 20 07:55 AM -- 08:00 AM (PDT) @ None

Graph Neural Networks (GNNs) have achieved remarkable performance on graph-based tasks. The key idea for GNNs is to obtain informative representation through aggregating information from local neighborhoods. However, it remains an open question whether the neighborhood information is adequately aggregated for learning effective node representations, especially for nodes with few neighbors. To address this, we propose a simple and efficient data augmentation strategy, local augmentation, to learn the conditional distribution of the node representations of each node's connected neighbors given their own representations and enhance GNN's expressive power with generated samples. Specifically, the well-learned conditional distribution contains generalized information about the local neighborhood and boosts the aggregation procedure of GNNs. Local augmentation is a general framework that can be applied to any GNN model in a plug-and-play manner. It extracts feature vectors associated with each node from the above distribution as additional input for the backbone model at each training epoch. Extensive experiments and analyses show that local augmentation consistently yields performance improvement when applied to various GNN architectures across a diverse set of benchmarks. For example, experiments show that plugging in local augmentation to GCN and GAT improves by an average of 7.2\% on three standard benchmark citation networks (cora, citeseer, and pubmed). Besides, our experimental results on large graphs (OGB) show that our model outperforms other feature/topology level augmentation models on node classification tasks.

Author Information

Songtao Liu (The Pennsylvania State University)
Rex (Zhitao) Ying (Stanford University)
Hanze Dong (HKUST)
Lanqing Li (Tencent AI Lab)
Tingyang Xu (Tencent Holdings)
Yu Rong (Tencent AI Lab)
Peilin Zhao (Tencent AI Lab)
Junzhou Huang (University of Texas at Arlington / Tencent AI Lab)
Dinghao Wu (Pennsylvania State University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors