BMI 539B Deep Learning II

Credits: 3
This course will cover two areas of deep learning in which labeled data is not required: Deep Generative Models and Unsupervised Learning. Recent advances in generative models have made it possible to realistically model high-dimensional raw data such as natural images, audio waveforms and text corpora. Topics include energy-based models (e.g., restricted Boltzmann machines), autoencoders, variational autoencoders, generative adversarial networks, in addition to a brief overview on elements of Bayesian inference including Monte Carlo techniques (e.g., Gibbs sampling and Metropolis-Hasting) and variational inference. This course will cover the theoretical foundations of these topics as well as their newly enabled applications. Students will learn how to effectively train deep models through unsupervised techniques, and will enable them to employ deep models in their research problems. The course will also draw from applications in speech and language processing. This is a second course in the sequence of “deep learning” topics and only those who have previously taken the “Deep Learning I” are encouraged to take this class. The course will be a combination of Introductory lectures, a few toy examples in Python, reading discussions in which students will take turns presenting papers and will be responsible for up to 2 papers. The course also requires a final project of interest to students chosen in consultation with the instructor. The project requires a written report and a final presentation.

Graded: A-F
May be taken only once for credit
Also offered as: BMI 639B