Instructor

Prof. Abir De

Semester

Autumn ‘21

Course Difficulty

I would consider this as a moderately difficult courses, on the basis of the mathematical rigor expected. The content is developed from a lot of recent research papers, thus there aren’t a lot of resources other than the class notes and research papers to fall back on. The project expectations are quite different based on the kind of project you choose (research/development), but the exams are difficult and require you to think on the spot rather than mug up stuff.

Time Commitment Required

The lectures if properly followed, are decent enough to develop a proper understanding of the content. A couple hours other than the lectures per week are good enough to stay on track. Towards the end, there were assignments, paper reviews and the final project of course, thus requiring more time (say 4-5 hours a week)

Grading Policy and Statistics

Decent, pretty relaxed I’d say. The absolute scores weren’t very high and the relative grading scheme was accordingly adjusted, not disclosed though. 6 AAs and 5 ABs in a batch of 21

Attendance Policy

None :)

Pre-requisites

Basics of graph theory (not mandated), basic course on machine learning (mandatory), good grasp on linear algebra, probability and optimization helps.

Evaluation Scheme

Total marks are calculated based on this formula.
Midsem - 20%
Endsem - 40%
Quiz (1) - 10%
Project, coding assignment and paper reviews - 30%

Topics Covered in the Course

Heuristic based as well as deep models for learning on graphs

Teaching Style

Live lectures with online tools for writing. Notes shared later with the students, frequent discussion sessions held to clarify doubts.

Tutorials/Assignments/Projects

No tuts, there was an assignment which was pretty basic, just to familiarize us with the use of graphical models on some standard datasets. Paper reviews required us to read and critically review any three papers out of the four choices given. The final project was to be selected from a pool of projects that the professor shared with the class. There were different kinds of projects with some requiring development, others demanding research. We were free to choose and work on any one of them in teams of at max 3. It served as a nice first project in the domain, interested peeps were open to continue working on the project after the course as well.

Feedback on Exams

Difficult! With most questions being open ended, the exams were difficult to score. The questions expected comfort with probability and algebra.

Motivation for taking this course

Wanted to explore learning approaches on graphical data, seeing their widespread applications, Netflix recommendations for one.

Course Importance

Helpful first course about learning approaches applied to graphs

How strongly would I recommend this course?

This is another focused course on a very particular topic, not another general ML course. If you’re interested in the domain of ML, this course offers an interesting insight into some very unique problems and building solutions around them.

When to take this course?

Took it in my 7th semester. The course can be taken once you’re comfortable with probability, linear algebra, machine learning and mathematical rigor in general. Decide for yourself, 5th sem might also be fine. Also the course is moderately heavy with not a lot of demands till midsems, but becomes a little more heavy from the evaluation perspective after midsems.

Going Forward

Research projects under the same prof is one good option. Anyway an interesting tool to have in your learning basket. Might be useful in some other projects too with similar settings.

Review By: Gagan Jain