Adaptive Multi-layer Contrastive Graph Neural Networks

Inspired by recent success of graph contrastive learning methods, we propose a self-supervised framework for Graph Neural Networks (GNNs) named Adaptive Multi-layer Contrastive (AMC-GNN). Specifically, AMC-GNN generates different views through data augmentation and compares the output embeddings at layers neural network encoders to obtain feature representations downstream tasks. Meanwhile, learns importance weights adaptively attention mechanism, an auxiliary encoder is adopted train better. The accuracy improved maximizing representation’s consistency positive pairs in intermediate final embedding space. Experiments on node classification link prediction demonstrate that outperforms state-of-the-art methods even sometimes supervised methods.