Zexi Huang successfully passed her MAE today on "Learning Node Representations for Information-rich Graphs".
Committee: Ambuj Singh (Chair), Yu-Xiang Wang, Noah Friedkin
Graphs are a ubiquitous data structure for encoding rich relational information in various domains, be it social networks, recommendation systems, protein-protein interactions, and many others. Node representation learning enables the application of classical machine learning algorithms for high-dimensional data to graph-based downstream tasks, such as link prediction, node classification, and community detection.
In this talk, I will start with an overview of machine learning on information-rich graphs. Then, I will review embedding methods based on random-walks and generalize them for graphs with multiple structural scales. Next, I will present representation learning on signed graphs and its application in characterizing polarization and sign prediction. I will then demonstrate how graph neural networks, the most popular paradigm for attributed graphs embedding, can be improved for sparsely labeled graphs. Finally, I will briefly discuss embedding for other types of graphs and propose some future research directions.