Special Course in Machine Learning: Probabilistic Deep Learning
The aim of the seminar course is to learn enough internal details of popular machine learning frameworks to be able to build probabilistic prediction models and use MCMC techniques to get samples from the posterior or complicated prior. As an added benefit you will learn also how to build standard neural networks with custom architecture.
There are two ecosystems on can use:
- TensorFlow and TensorFlow Probability (Google)
- PyTorch (Facebook) and Pyro (Uber)
In the first seminars we cover the technical differences between these alternatives and then pick one of them as the base.
Each topic will be covered as a seminar presentation that is illustrated by Python code that will be made public in the course repository. To complete the course, you have to do a presentation and tidy up the corresponding example code.
Most of the course materials will be in the Github repository including the links to interesting materials.
Prerequisites. There are no official prerequisites. However, you must have a resonable background in math and deep learning.
- Basic calculus
- Machine learning
- Programming
Materials. All materials and sample code will be in the Github repository
Requirements to pass the course: TBA
Contacts. The course is held by Sven Laur. Mail me to the university mail address or drop by to my office Delta 3108 but contact me before to schedule a meeting.