GrAPL 2021 Keynote 2: Label Propagation and Graph Neural Networks
2021
Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. Semi-supervised learning on graphs is a widely applicable problem in network science and machine learning. Two standard algorithms - label propagation and graph neural networks - both operate by repeatedly passing information along edges, the former by passing labels and the latter by passing node features, modulated by neural networks. These two types of algorithms have largely developed separately, and there is little understanding about their relationship and how the approaches can be meaningfully combined. In this talk, I will present some probabilistic models that unify these algorithms, showing how label propagation and graph neural network ideas are naturally connected and how this leads to algorithms that can use both effectively. The talk will also discuss computational and machine learning tradeoffs of complex, highly expressive models that are expensive to train and difficult to implement, compared to simpler, less expressive models that run faster, are easy to implement, and offer more opportunities for parallelism.