Basic Information
- Course Code: CS 768
- Course Name: Learning with Graphs
- Course Offered In: 2023-‘24
- Semester Season: Autumn
- Instructors: Prof. Abir De
- Prerequisites: Intro to ML (DS303/CS419 etc)
- Difficulty (1 being easy and 5 being tough): 4
Course Content
A lot of topics are covered in a short time. Sir keeps shifting from one topic to the next one very quickly, so attending classes is a must. The course contents mainly cover 1) Differences between training on graphs and other data structures 2) Different paradigms of learning on graphs : Link Prediction, Node/Graph Classification etc. 3) Some commonly applied strategies while learning on graphs : Message passing, Locality Sensitive Hashing etc 4) Miscellaneous concepts relating to graphs (information diffusion, influence maximization)
All of the above are covered in all its mathematical detail, and the instructor very often spends quite a lot of time on one problem if he is not able to get a good enough answer from the students.
The contents can vary quite a lot depending on the inclination and level of the attending students.
Feedback on Lectures
The lectures are very math intensive, and a decent level of concentration is required throughout the lecture to be able to follow along. The prof. expects an intermediate level of understanding of a wide variety of ML concepts, and thus tends to skip over the easier to understand details during his explanations. However, the course offers a very nice exposure to research in graph ML, as well as teaches one how to think and come up with solutions to problems. You will be required to go through a plethora of research papers during the course, and will be exposed to a bunch of methods and ideologies in ML.
Feedback on Evaluations
The course consisted of 3 quizzes (with a coding component in one), out of which best 2 were selected. Besides, the weightages were as follows : 1) Project : 40% Mainly reproducing and extending the work from one research paper. The paper is chosen from a list provided by the instructor 2) Endsem : 40% 3) Quizzes : 15% 4) Scribes : 5%
The grading is very lenient if one attends all lectures. The questions are asked from one or more research papers, and asks the student to complete derivations (steps are indicated) and/or make sense of the solutions. They are usually solvable if one follows the class discussions.
Study Material and Resources
Is provided by the instructor. No single text book is recommended/sufficient for the course.
Final Takeaway
This is a must-take for people who want to get into research. For the DD students, it might be better to take it in the 5th year, when the course load is lesser, and thus more time can be devoted to take in the contents of this course.