Instructor

Prof. Harish Guruprasad Ramaswamy

Semester

Autumn ‘21

Course Difficulty

The course certainly demanded mathematical rigour to some extent as well as the ability to grasp concepts from research papers. A basic understanding of (as well as some coursework in) machine learning was expected before taking the course.

Time Commitment Required

The course had discrete peaks as far as time commitment is concerned. The paper presentations and the final project demanded time towards the later half of the course. The quizzes were very easy if the lecture content was properly followed. But apart from the evaluation, paper reading was expected throughout the semester so one can expect spending around 3 hours per week apart from the usual lecture hours if they wish to follow the course religiously. The discrete peaks I talked about earlier demanded around 7-8 hours each for the paper presentations and around 20-25 hours of work for the project. This might vary from person to person but just to give a rough idea.

Grading Policy and Statistics

Somewhat chiller than absolute but not too chill either. 4 AAs and 9 ABs in a batch of 29.

Attendance Policy

None :)

Pre-requisites

Coursework in basic Machine Learning was mandatory. Good knowledge of linear algebra, calculus and optimization helps in appreciating the course content better.

Evaluation Scheme

2 quizzes - 5% each
2 paper presentations - 30% each
course project - 30% + 5% bonus for extending the idea of a research paper and presenting extra results

Topics Covered in the Course

Fairness
#Notions of fairness
#Constrained classification
#Algorithmic vs data unfairness

Explainability/Interpretability/Visualization
#Visualizations of neural nets
#Interpretable Machine Learning

Other Topics
#Privacy (differential privacy, information theory)
#Distributed/federated learning
#Ethics in AI

Teaching Style

Live lectures, the professor used online writing tools to explain concepts, these notes were later shared with the class along with the recordings of the lectures. The latter half of the lecture hours were dedicated to paper presentations by students.

Tutorials/Assignments/Projects

Being a small class, the mode of communicating was directing messaging the prof for taking any feedback or asking doubts related to the paper presentations and projects. These evaluations helped concentrate on a particular problem and go about solving it in an organised manner.

Feedback on Exams

There were no exams other than the two quizzes. They were super easy given that the lecture content is properly followed.

Motivation for taking this course

The course offered interesting recent topics in the domain of Machine Learning, primarily around Ethics in AI and Interpretibility of Deep Networks. I wanted to explore these aspects in a systematic manner.

Course Importance

Serves as a good starter for the abovementioned domains, the discussed ideas can be translated into research work later on in the field.

How strongly would I recommend this course?

Would only recommend if you’re interested in these specific topics (Ethics in AI and Explainable AI) and are willing to dedicate sufficient time in understanding more about them. The course has a very narrow and dedicated scope, and anyone looking for a general course around machine learning should stay away.

When to take this course?

Took it in my 7th semester. Given that usually people interested in ML do an introductory course by their 4th semester, taking this in 5th semester should also be fine.

Going Forward

Possibly research projects in these domains

References Used

Check the papers listed here - https://sites.google.com/site/harishguruprasad/teaching/topics-in-ml-iitb-aug-2021

Review By: Gagan Jain