本文提出基于知识蒸馏的标准化 GNN Self-Distillation (GNN-SD) 框架来替代传统 GNN 的训练过程。GNN-SD 基于本文提出的用于量化图嵌入非光滑性的 neighborhood discrepancy rate (NDR) 指标进行实现。同时作者提出 adaptive discrepancy retaining (ADR) 正则化器来增强知识的可转移性。实验证明了该方法的有效性。 GNN...
On Self-Distilling Graph Neural Network.doi:10.24963/IJCAI.2021/314Yuzhao ChenYatao BianXi XiaoYu RongTingyang XuJunzhou HuangInternational Joint Conferences on Artificial Intelligence Organization
On Self-Distilling Graph Neural Network, Yuzhao Chen, Yatao Bian, Xi Xiao, Yu Rong, Tingyang Xu, Junzhou Huang Learning Robust Node Representations on Graphs, Xu Chen, Ya Zhang, Ivor Tsang, and Yuangang Pan Recurrent Event Network: Autoregressive Structure Inference over Temporal Knowledge Gr...
Dual Self-Paced Graph Convolutional Network: Towards Reducing Attribute Distortions Induced by Topology. IJCAI 2019. paper Liang Yang, Zhiyang Chen, Junhua Gu, Yuanfang Guo. Bayesian graph convolutional neural networks for semi-supervised classification. AAAI 2019. paper Yingxue Zhang, Soumyasundar Pal...
GNN: graph neural network Contributed by Jie Zhou, Ganqu Cui, Zhengyan Zhang and Yushi Bai. 来源:THU
represents neighborhood discrepancy as local structures and focuses on distilling them from shallow GNN layers to deep ones in order to preserve the non-smoothness of the embedded graph on deep GNNs. 结构。图结构描述了图的元素(如节点和边)之间的连接和关系,在图数据建模中起着至关重要的作用。因此,...
Dual Self-Paced Graph Convolutional Network: Towards Reducing Attribute Distortions Induced by Topology. IJCAI 2019. paper Liang Yang, Zhiyang Chen, Junhua Gu, Yuanfang Guo. Bayesian graph convolutional neural networks for semi-supervised classification. AAAI 2019. paper Yingxue Zhang, Soumyasundar Pal...
Previous KD often focused solely on distilling from the logits layer or the feature intermediate layers, which may limit the comprehensive learning of the student network. Additionally, the imbalance between the foreground and background also affects the performance of the model. To address these ...
Although deep reinforcement learning has become a promising machine learning approach for sequential decision-making problems, it is still not mature enoug
类似的,ABUSNAINA等人[61]引入了潜在邻域图(Latent Neighborhood Graph, LNG),将对抗样本检测问题转化为图分类问题,通过图神经网络(Graph Neural Network, GNN)优化LNG节点连接参数,区分良性和对抗样本。 针对单一检测方法效果不佳的问题,徐东伟等人[62]提出了一种结合决策边界敏感性和小波变换重构的对抗样本检测方法。