Artun Bayer

Artun Bayer, Arindam Chowdhury, Santiago Segarra

Graph neural networks (GNNs) have achieved superior performance on node classification tasks in the last few years. Commonly, this is framed in a transductive semi-supervised learning setup wherein the entire graph, including the target nodes to be labeled, is available for training. Driven in part by scalability, recent works have focused on the inductive case where only the labeled portion of a graph is available for training. In this context, our current work considers a challenging inductive setting where a set of labeled graphs are available for training while the unlabeled target graph is completely separate, i.e., there are no connections between labeled and unlabeled nodes. Under the implicit assumption that the testing and training graphs come from similar distributions, our goal is to develop a labeling function that generalizes to unobserved connectivity structures. To that end, we employ a graph neural tangent kernel (GNTK) that corresponds to infinitely wide GNNs to find correspondences between nodes in different graphs based on both the topology and the node features. We augment the capabilities of the GNTK with residual connections and empirically illustrate its performance gains on standard benchmarks.

Cameron R. Wolfe, Jingkang Yang, Arindam Chowdhury et al.

The graph convolutional network (GCN) is a go-to solution for machine learning on graphs, but its training is notoriously difficult to scale both in terms of graph size and the number of model parameters. Although some work has explored training on large-scale graphs (e.g., GraphSAGE, ClusterGCN, etc.), we pioneer efficient training of large-scale GCN models (i.e., ultra-wide, overparameterized models) with the proposal of a novel, distributed training framework. Our proposed training methodology, called GIST, disjointly partitions the parameters of a GCN model into several, smaller sub-GCNs that are trained independently and in parallel. In addition to being compatible with all GCN architectures and existing sampling techniques for efficient GCN training, GIST i) improves model performance, ii) scales to training on arbitrarily large graphs, iii) decreases wall-clock training time, and iv) enables the training of markedly overparameterized GCN models. Remarkably, with GIST, we train an astonishgly-wide 32,768-dimensional GraphSAGE model, which exceeds the capacity of a single GPU by a factor of 8x, to SOTA performance on the Amazon2M dataset.