CSCI 6995G - Graph Representation Learning in Digital Media

Complex graph-structured data such as social networks, information networks, and knowledge graphs have fueled many applications in digital (social) media ranging from recommender systems to link predictions to automated question answering. Graph representation learning is a growing branch of machine learning which focuses on learning representations on graph-structured data. Recent advances in this area include graph embeddings, graph convolutional neural networks, and neural message-passing networks. These advances have revolutionized many new applications in digital media. This course provides the foundational knowledge on graph representation learning. Students learn various aspects of graph representation learning including node embedding, graph embedding, label prediction on graphs, link prediction, graph convolutional networks, graph neural networks, deep generative models for graphs, etc. They gain hands-on experience with open-source libraries (e.g., PyTorch, PyTorch Geometric, DeepSNAP, GraphGym, etc.) and large-scale graph-structured datasets. They also learn the theory behind the state-of-the-art technology for graph representation learning in digital media applications such as recommender systems and question answering.
Credit hours: 3