Graph Neural Networks - Training on very large graphs

Ever-growing size of social interactions, especially via on-line channels, creates a need for processing big datasets in an efficient way. Thanks to the verbosity level of Python combines with C speed Julia is a perfect tool to implement large graph processing algorithms. The purpose of this project is to enrich GraphNeuralNetworks.jl library with algorithms for processing large graphs based on the literature. First part of the project will focus on adding GraphSAGE in basic version and its variations (e.g. PinSAGE and UberEats inspired), including different sampling techniques and variety of aggregation operators. Then, less conventional approaches will be implemented, such as Cluster-GCN and algorithms based on linear algebra optimization. Finally, the project will be completed by creating tutorials for the new features.