ES 661 IIT Gandhinagar

Home Lectures Assessment Assignments Interactive apps Slack

Probabilistic Machine Learning

ES 661 · IIT Gandhinagar · Semester I, 2023–24

ES 661 is a course on probabilistic machine learning: Bayesian inference, estimation, sampling, approximate inference, probabilistic models, and their use in Bayesian optimisation and active learning. Lectures link their recordings, slides, cheat sheets and notebooks.

Instructor

Prof. Nipun Batra

Email

nipun.batra@iitgn.ac.in

Lectures

Monday 3:30–4:50 pm and Thursday 2:00–3:20 pm IST

Teaching assistants

4 TAs

Discussion

Course Slack

Course record

This offering ran from August to November 2023. Its lectures, recordings, notebooks and assignments remain available.

Lectures

25

Recordings

23

Cheat sheets

15

Assignments

4

Interactive apps

17

All dates

Lectures

• 1–4Probability and likelihood4 lectures
• 5–8Estimation and Bayesian regression4 lectures
• 9–14Sampling and MCMC6 lectures
• 15–16Uncertainty in deep learning2 lectures
• 17–20Hierarchical models, meta-learning and self-supervision4 lectures
• 21–22Active learning and Bayesian optimisation2 lectures
• 23–25Variational inference3 lectures

Lecture playlist All 25 lectures

Questions

How is the course graded?

Three quizzes (36%), four assignments (32%), a group project (27%) and an interactive article (5%). See Grading.

When are the quizzes?

There are three quizzes of 12% each, held in the exam 1, 2 and 3 slots.

How do the assignments work?

Each assignment is a single Jupyter notebook done in a group, using PyTorch and torch.distributions, with about a week to complete it. Each is followed by a viva and graded on correctness of the code (50%), code quality (25%) and presentation and conceptual understanding (25%).

Is there a project?

Yes, in groups of three to five students. It has a first presentation, a final presentation and a final report. See Project.

What is the interactive article?

A Streamlit app or similar interactive article that explains a concept from the course. See the course's interactive apps.

Ask anything else on the course Slack.

Main topics

1. Bayesian inference
2. Estimation: maximum likelihood, maximum a posteriori, full Bayesian
3. Sampling: rejection sampling, Monte Carlo, specific sampling techniques such as Box–Muller
4. Approximate inference: variational inference, Markov chain Monte Carlo, Laplace approximation
5. Models: Bayesian linear and logistic regression, Bayesian neural networks, Gaussian processes, probabilistic PCA
6. Applications: Bayesian optimisation, active learning

Teaching assistants

• Zeel B Patel
• Sarth Dubey
• Madhav Kanda
• Haikoo Khandor