ACMS 80770: Deep Learning with GraphsInstructor: Navid Shervani-Tabar ([email protected]), 203F Crowley Hall
TR 3:30p-4:45p, 231 DeBartolo Hall Fall 2022 From the emergence of the world wide web to the rise of social media, graphs have been the cornerstone of many revolutionizing advances in the past two decades. This results from graphs’ natural ability to represent the interactions in multi-component systems, such as users in a social network or the amino acids in a protein folding problem. The increased importance of graphs in our daily lives has rendered them a target for machine learning techniques. With AI’s emergence, deep learning methods are becoming the mainstream prediction approach on graphs. This course introduces the fundamental concepts of data analysis and learning on graphs, with an emphasis on generalizing deep neural models to this non-Euclidean domain, with applications to drug discovery, computational biology, recommender systems, and social network analysis. It covers complex data representation with graphs, graph structure and analysis measures, feature extraction from graphs for machine learning models, representation learning approaches, methods for deep learning on graphs, generative graph models, spectral graph theory, and dynamic graphs. Throughout the course, students will build their own deep learning models that will be used to study node, edge, and graph-level learning and investigate benchmark datasets such as citation network. In addition, the course makes use of open source libraries to explore more advanced approaches, validate their own implemented models, and gain experience with software commonly used in science and engineering.
|
Lectures
|