Original title: Domain Adaptive Graph Classification
Authors: Siyang Luo, Ziyi Jiang, Zhenghan Chen, Xiaoxuan Liang
In the article, the authors discuss the challenges that arise in the field of graph neural networks (GNNs) due to the reliance on task-specific labels. They highlight that obtaining these labels can be a difficult and time-consuming process. To address this issue, previous research has focused on unsupervised domain adaptation, which involves utilizing labeled source graphs to improve learning for target data. However, there is still a significant obstacle in simultaneously exploring graph topology and reducing domain disparities.
To overcome these challenges, the authors propose a new method called Dual Adversarial Graph Representation Learning (DAGRL). This approach incorporates a dual-pronged structure that includes a graph convolutional network branch and a graph kernel branch. By employing both branches, they are able to capture graph semantics from both implicit and explicit perspectives.
Additionally, DAGRL introduces adaptive perturbations into the dual branches, which helps align the source and target distribution to address domain discrepancies. The effectiveness of this method is demonstrated through extensive experiments on various graph classification datasets.
Original article: https://arxiv.org/abs/2312.13536