Extract the Knowledge of Graph Neural Networks and Go Beyond it: An Effective Knowledge Distillation Framework

4 Mar 2021  ·  Cheng Yang, Jiawei Liu, Chuan Shi ·

Semi-supervised learning on graphs is an important problem in the machine learning area. In recent years, state-of-the-art classification methods based on graph neural networks (GNNs) have shown their superiority over traditional ones such as label propagation. However, the sophisticated architectures of these neural models will lead to a complex prediction mechanism, which could not make full use of valuable prior knowledge lying in the data, e.g., structurally correlated nodes tend to have the same class. In this paper, we propose a framework based on knowledge distillation to address the above issues. Our framework extracts the knowledge of an arbitrary learned GNN model (teacher model), and injects it into a well-designed student model. The student model is built with two simple prediction mechanisms, i.e., label propagation and feature transformation, which naturally preserves structure-based and feature-based prior knowledge, respectively. In specific, we design the student model as a trainable combination of parameterized label propagation and feature transformation modules. As a result, the learned student can benefit from both prior knowledge and the knowledge in GNN teachers for more effective predictions. Moreover, the learned student model has a more interpretable prediction process than GNNs. We conduct experiments on five public benchmark datasets and employ seven GNN models including GCN, GAT, APPNP, SAGE, SGC, GCNII and GLP as the teacher models. Experimental results show that the learned student model can consistently outperform its corresponding teacher model by 1.4% - 4.7% on average. Code and data are available at https://github.com/BUPT-GAMMA/CPF

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Node Classification AMZ Computers CPF-ind-GAT Accuracy 85.5% # 1
Node Classification AMZ Photo CPF-ind-GAT Accuracy 94.10% # 1
Node Classification CiteSeer with Public Split: fixed 20 nodes per class CPF-tra-APPNP Accuracy 74.6% # 3
Node Classification Cora (0.5%) CPF-ind_APPNP Accuracy 77.3% # 1
Node Classification Cora (1%) CPF-ind-APPNP Accuracy 80.24% # 2
Node Classification Cora (3%) CPF-tra-GCNII Accuracy 84.18% # 1
Node Classification Cora: fixed 10 node per class CPF-tra-GCNII Accuracy 84.1% # 1
Node Classification Cora: fixed 5 node per class CPF-tra-APPNP Accuracy 80.26% # 1
Node Classification Cora with Public Split: fixed 20 nodes per class CPF-ind-APPNP Accuracy 85.3% # 3
Node Classification PubMed with Public Split: fixed 20 nodes per class CPF-tra-GCNII Accuracy 83.20% # 2

Methods