IFT 6760B A25 - Probabilistic inference with GFlowNets
Course description
Generative flow networks, also known as GFlowNets or simply GFN, are a class of generative machine learning models that perform amortised probabilistic inference. This course will cover the fundamental aspects of GFlowNets, starting from a motivation and introduction to the method, and progressing towards more advanced concepts and applications, as well as the connection to other generative models and probabilistic inference methods. The course will combine theory with project work, with a special emphasis on the application of GFlowNets for scientific discovery.
Les réseaux de flux génératifs, également connus sous le nom de GFlowNets ou simplement GFN, sont une classe de modèles d’apprentissage automatique génératifs qui effectuent une inférence probabiliste amortie. Ce cours couvrira les aspects fondamentaux de GFlowNets, à partir d’une motivation et d’une introduction à la méthode, en progressant vers des concepts et applications plus avancés, ainsi que le lien avec d’autres modèles génératifs et méthodes d’inférence probabiliste. Le cours combinera la théorie avec le travail de projet, avec un accent particulier sur l’application des GFlowNets pour la découverte scientifique.
Course outline
Coming soon
Evaluation criteria
Coming soon
Prerequisites
As a prerequisite to register for this course, students must have successfully completed the following courses:
- Introduction à la science des données (IFT 3700)
- Fundamentals of machine learning (IFT 3395/6390).
Additionally, it is recommended to have taken (or take in parallel) the following courses:
- Representation learning (IFT 6135)
- Probabilistic graphical models (IFT 6269)
Comme exigences d’inscription à ce cours, les étudiant.e.s doivent avoir réussi :
- Introduction à la science des données (IFT 3700)
- Fondements de l’apprentissage machine (IFT 3395/6390).
En outre, il est recommandé d’avoir suivi (ou de suivre en parallèle) les cours suivants :
- Apprentissage de représentations (IFT 6135)
- Modèles graphiques probabilistes et apprentissage (IFT 6269)
Why such prerequisites?
In order to cover the specifics of generative flow networks (GFlowNets), we have to depart from a position of familiarity with the fundamental concepts of machine learning as well as deep learning. Additionally, we will be using the language of probability, statistics, linear algebra and calculus. A good reference for the contents that are expected to be familiar with is Part I of the Deep Learning book, by Goodfellow, Bengio and Courville. Finally, the course includes project work, which requires familiarity with Python and the typical machine learning libraries, such numpy, pandas, PyTorch, etc. You can check the resources section below for additional references.
Useful links
Session d’automne 2025
Les cours ont lieu :
- Les lundis, 10h30–12h30 (ET)
- Les jeudis, 10h30–12h30 (ET)
Resources
- Abu-Mostafa, Y. S., Magdon-Ismail, M., & Lin, H. T. (2012). Learning from data. AMLBook.
- Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press.
- Anish, Jose, Jon (last visit on May 5th, 2025). The Missing Semester of Your CS Education. CSAIL MIT.
- Linux introduction for Windows and Mac users. Compute Canada wiki.
- Python tutorial: An informal introduction to Python. www.python.org.
- PyTorch tutorials. pytorch.org.
- Deep Learning Tutorials. Neuromatch Deep Learning course.