Browsing by Subject "Bayesian inference"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    How Does Bayesian Noisy Self-Supervision Defend Graph Convolutional Networks?
    (Springer, 2022-08) Zhuang, Jun; Al Hasan, Mohammad; Computer and Information Science, School of Science
    In recent years, it has been shown that, compared to other contemporary machine learning models, graph convolutional networks (GCNs) achieve superior performance on the node classification task. However, two potential issues threaten the robustness of GCNs, label scarcity and adversarial attacks. .Intensive studies aim to strengthen the robustness of GCNs from three perspectives, the self-supervision-based method, the adversarial-based method, and the detection-based method. Yet, all of the above-mentioned methods can barely handle both issues simultaneously. In this paper, we hypothesize noisy supervision as a kind of self-supervised learning method and then propose a novel Bayesian graph noisy self-supervision model, namely GraphNS, to address both issues. Extensive experiments demonstrate that GraphNS can significantly enhance node classification against both label scarcity and adversarial attacks. This enhancement proves to be generalized over four classic GCNs and is superior to the competing methods across six public graph datasets.
  • Thumbnail Image
    Item
    Robust Node Classification on Graphs: Jointly from Bayesian Label Transition and Topology-based Label Propagation
    (ACM, 2022-10-17) Zhuang, Jun; Al Hasan, Mohammad; Computer and Information Science, School of Science
    Node classification using Graph Neural Networks (GNNs) has been widely applied in various real-world scenarios. However, in recent years, compelling evidence emerges that the performance of GNN-based node classification may deteriorate substantially by topological perturbation, such as random connections or adversarial attacks. Various solutions, such as topological denoising methods and mechanism design methods, have been proposed to develop robust GNN-based node classifiers but none of these works can fully address the problems related to topological perturbations. Recently, the Bayesian label transition model is proposed to tackle this issue but its slow convergence may lead to inferior performance. In this work, we propose a new label inference model, namely LInDT, which integrates both Bayesian label transition and topology-based label propagation for improving the robustness of GNNs against topological perturbations. LInDT is superior to existing label transition methods as it improves the label prediction of uncertain nodes by utilizing neighborhood-based label propagation leading to better convergence of label inference. Besides, LIndT adopts asymmetric Dirichlet distribution as a prior, which also helps it to improve label inference. Extensive experiments on five graph datasets demonstrate the superiority of LInDT for GNN-based node classification under three scenarios of topological perturbations.