|Instructor:||A. Erdem Sariyuce (erdem AT buffalo.edu)|
|Class hours:||W 6-8.30, Davis 338A (if not online)|
|Office hours:||W 12-2, Davis 323|
Graphs are everywhere. Their scale, rate of change, and the irregular nature pose many new challenges. Deep learning has been shown to be successful in a number of domains, ranging from images to natural language processing. However, applying deep learning to the ubiquitous graph data is non-trivial because of the unique characteristics. This seminar course covers recent papers in the last few years about deep learning on graphs. We will consider Graph Recurrent Neural Networks, Graph Reinforcement Learning, Graph Adversarial Methods, Graph Convolutional Networks, and Graph Autoencoders. Students will learn the literature on deep learning on graphs, understand the state-of-the-art algorithms on various problems, and be familiar with the recent trends.
It is assumed that students have a solid background on discrete mathematics and algorithms. Basic research skills like paper reading, critical thinking, problem solving, report writing, communication, and presentation are important as well.
The final grade is S/U and 75% score is needed for an S. Regarding the literature survey (for students taking 3 credits), topic will be decided with instructor.
Each student picks 1-2 papers from the reading list and present. Tentative schedule is below. A presentation is expected to be an hour long. Each week, all the students will read the papers of the week before class and will ask a unique question on Piazza (except the presenter) to facilitate a class discussion. Questions should be open-ended and provide ground for class discussions, i.e., 'can you explain alg 1?' is not that kind of question. Questions should be posted to Piazza by Monday night, 11.59 pm EST.