Poster Presentation: Towards Decentralized GNNs
Graph Neural Networks (GNNs) achieve state-of-the-art results in most graph representation learning benchmarks. However, compared to other deep learning models, their structure makes them hard to decentralize. Yet, decentralization is an important tool to achieve large-scale, data-private machine learning. In this work, we show how layer-wise, self-supervised learning may be used to train deep GNNs on a decentralized graph, where each node represents a separate computing device.