Columbia STAT 8201(1) Spring 2019 John Cunningham

Course Description

Deep generative models aim to combine the interpretable representations and quantified uncertainty offered by probabilistic models, with the flexibility and scalable learning of deep neural networks. This research area -- which includes variational autoencoders, generative adversarial networks, and more -- is one of the most exciting and rapidly evolving fields of statistical machine learning. This seminar course will explore the cutting edge of deep generative models. This is a PhD-level seminar course, and will involve a major project by the student (or group of two students). Students have substantial freedom in choosing these projects, from creating novel research-grade methods, to contributing to open source machine learning projects, to analyzing data of interest, to exploring a theoretical topic. Projects that connect to the student's ongoing research agenda are particularly welcome. Proposals and reports will be done in the format and style of top machine learning conferences, with the intention that some projects will be developed into papers at those venues.