Adaptive Graph Convolutional Neural Networks
This work addresses a key bottleneck in graph-based machine learning for domains like molecular data, point clouds, and social networks, offering a more adaptable approach to handle varying graph structures.
The paper tackled the problem of graph convolutional neural networks (Graph CNNs) being limited to fixed and shared graph structures, which do not align with real-world data where graph structures vary in size and connectivity. It proposed a generalized and flexible Graph CNN that learns task-driven adaptive graphs for each input, resulting in superior performance improvements in convergence speed and predictive accuracy across nine graph-structured datasets.
Graph Convolutional Neural Networks (Graph CNNs) are generalizations of classical CNNs to handle graph data such as molecular data, point could and social networks. Current filters in graph CNNs are built for fixed and shared graph structure. However, for most real data, the graph structures varies in both size and connectivity. The paper proposes a generalized and flexible graph CNN taking data of arbitrary graph structure as input. In that way a task-driven adaptive graph is learned for each graph data while training. To efficiently learn the graph, a distance metric learning is proposed. Extensive experiments on nine graph-structured datasets have demonstrated the superior performance improvement on both convergence speed and predictive accuracy.