Mert Kosan successfully passed his Major Area Exam (MAE) today on “Interpretable Representation Learning for Attributed Graphs”.
Committee: Ambuj Singh (Chair), Xifeng Yan, Francesco Bullo
Graph data are effective for showing relationships between entities in a variety of domains including but not limited to communication, social, and interaction networks. Representation learning makes graph data easier to use for graph-level tasks such as classification and event detection. Furthermore, explainable models are essential for detecting patterns in the data.
During the talk, I will motivate why graphs and representation learning are useful for machine learning, and why we should work on interpretable models. Then, I will review the recent literature around graph representation learning, interpretable models, and inductive-transductive tradeoff along with their related applications including our ongoing work; event detection of dynamic graphs and important node detection for graph classification using bilevel optimization. I will also talk about how machine learning can be adapted to decision making under uncertainty. Lastly, I will discuss future research directions.