From Local Structures to Size Generalization in Graph Neural Networks

           Gilad Yehudai 1 Ethan Fetaya 2 Eli Meirom 1 Gal Chechik 1 2 Haggai Maron 1

             Abstract
    Graph neural networks (GNNs) can process
    graphs of different sizes, but their ability to gen-
    eralize across sizes, specifically from small to
    large graphs, is still not well understood. In
    this paper, we identify an important type of data
    where generalization from small to large graphs
    is c ...