Special Course in Machine Learning: Geometric Deep Learning
Geometric Deep Learning (GDL) is an effort to exploit the geometric structure of datasets to achieve efficient representation learning. But GDL also proposes a fundamental principle (symmetry) as a framework to understand and classify the current (and future) zoo of diverse neural networks.
In this course, we are going to delve into geometric deep learning theory based on the public online course by Michael M. Bronstein, Joan Bruna, Taco Cohen, and Petar Velickovic: https://geometricdeeplearning.com/lectures/
Lecture 1: Introduction Lecture 2: High-Dimensional Learning Lecture 3: Geometric Priors I Lecture 4: Geometric Priors II Lecture 5: Graphs & Sets I Lecture 6: Graphs & Sets II Lecture 7: Grids Lecture 8: Groups Lecture 9: Geodesics & Manifolds Lecture 10: Sequences & Time Warping Lecture 11: Conclusions Tutorial 1: Graph Neural Networks Seminar 1: Geometric Deep Learning and Reinforcement Learning
This course requires a strong background in math and deep learning.
- A second course in Algebra
- Deep learning
The course consists of video lectures and tests/project. Each week we will meet to watch one lecture from https://geometricdeeplearning.com/lectures/ in which we will stop the video frequently to discuss. It will be possible to participate in the discussion/tests fully online via Zoom for those students who cannot attend physically.
The first lecture will take place on Friday, September 10th.
The class is held in Delta room 2049 on Fridays at 12:15-14. For each class one of the teachers/students (or team) will produce test questions for the video lecture of that week. NB! The test must be sent to email@example.com by Friday before the class for review!
The last 4 weeks are reserved for a team project. The idea of a project is to implement some concepts covered during the course.
To pass the course one has to:
- create one test and score more than 70% in all other tests,
- participate in a team project.
- https://geometricdeeplearning.com/lectures/ - we will mostly follow this course.
- https://arxiv.org/abs/2104.13478 - arXiv paper with most of the content of the lectures.