CS 791: Probabilistic Foundations of AI (Autumn 2025)


Lecture Schedule Slot 12, Mon,Thu 5:30pm--6:55pm.
Venue
Instructor: Sunita Sarawagi
TAs: Darshan Prabhu (PhD), Gaurav Kumar (MTech), Daksh Goyal (MTech), Ritesh Sur Chowdhury (MS by R), Aditya Neeraje (BTech3), Deeptanshu Malu (Btech4), Deevyanshu Malu (Btech4)
Email to reach all TAs and Instructor CS791@googlegroups.com.
For questions of general interest to the class, use moodle.
Instructor's office hours check here
Syllabus and week-wise calendar: Click here

Course description

This course will concentrate on modeling, generation, and prediction of multiple inter-dependent variables. The topics covered will span probabilistics graphical models (directed and undirected), inference methods like junction trees, belief propagation, and other approximate methods, MCMC sampling methods like Gibbs and Langevin, deep generative models like diffusion models, variational auto-encoders, GANs, Gaussian processes, causality, and other recent topics in machine learning. When appropriate the techniques will be linked to applications in NLP, speed recognition, vision, graphics, and sciences.

CS 791 was earlier called CS 726 (Advanced ML). The new name is to more faithfully reflect the contents of the course. Student who have already taken CS 726 should not take this course. It will be considered a duplicate.

Eligibility

The course is open to PhD, Masters, DD and BTech students provided they have taken an introductory course in IITB in machine learning (listed below) and obtained at least a BC grade in it.

Prerequisites

You are required to have taken a formal introductory ML course at IITB like CS 725 or CS 217 or CS 419 or DS 303 or equivalent courses in other departments. Online ML courses do not qualify as pre-requisites. The course assumes basic knowledge of probability, statistics, linear algebra, and numerical optimization. Chapters 2, 3, 4 of the Deep-learning book are a good place to refresh the necessary required background. Also, the course assumes basic background in machine learning, for example as covered in Chapter 5 of the Deep-learning book and deep learning, for example, as covered in Chapter 6 of the same book. Further, we will assume that students are familiar with CNNs, RNNs, and sequence to sequence learning with attention. Also, there will be programming assignments that will benefit from fluency in Python. A quiz will be administered within the course drop deadline to help you assess if you have the requisite background for the course. You may be asked to drop the course if you score below a threshold in that quiz.

Credit/Audit Requirements

Approximate credit structure

Audit students have to score more than 30% over all, spanning any of assignments, quizzes and mid/end sem exam. Also, audit students need to appear in at least 80% of the class quizzes.

Reading List

Primary text books for part of the course

Probabilistic Graphical Models: Principles and Techniques, by Daphne Koller and Nir Friedman, MIT Press, 2009.
The course calendar will provide links to other relevant papers and book chapters for specific topics.