Autumn 2021
Prof. Biplab Banerjee
Author: Ayush Sarraf
Pre-requisite courses: No hard prerequisite. Basic knowledge of Linear Algebra is expected. Having some very intro level ML familiarity (from random online course) will give you a slight advantage, and make the course chiller for you.
Pre-requisite skills: Good to know python as it is required during the kaggle competition and the project
Course Content:
- Recognition problems in computer vision and remote sensing
- Evolution of feature extraction and representations
- Neural Networks and CNNs, discussion in detail of Important CNN architectures (VGGNet, AlexNet, GoogleNet, ResNets, etc.), advanced CNN models (Bayesian CNN, Siamese/Triplet CNN)
- Multi-modal learning And many more!
Motivation to take up the course: This course gives a high level overview of most of the hottest topics in ML, leaving the student to explore more based on their interests. After taking this course, you will be able to understand the basic premise of most ML applications and approaches, and will also be able to think of ML-driven solutions to problems. Also, the evaluation scheme is very chill, and the grading is generous, so grading wise also the course is an ideal Institute Elective.
Evaluation Structure: 2 Paper Reviews - 20% (critical reviews of research papers) 1 Kaggle competition - 10% Course Project - 25% 2 Quizzes - 45%
Information about Projects/Assignments: The paper reviews were done in groups of 2 and was basically to summarize the content of the paper in 1-2 pages. The kaggle competition was typically based on training a multi class image classification task using CNN. The project could be done in groups of 3-5 and involved submitting a presentation and the codes on a topic of your choice related to the course.
Coding Assignments (Individual): Application of known algorithms covered in lectures. They were fairly straight forward and took around 4-5 hours to complete. These assignments helped in understanding the implementation of the theory taught in class.
Course Project (Group): Students were required to implement a research paper of their choice and demonstrate the results in the final presentation.
Quizzes/Midsem/Endsem papers Difficulty: 3/5
Difficulty level of Projects/Assignments: 2/5
Attendance Policy: No DX grade enforced
General funda: Try attending the lectures and also atleast submitting everything. Do not miss out on submitting any grading element as marks are easy to come.
Feedback on Lecture: The professor has regular lectures where new concepts are covered. A lot of content is covered in a lecture, and depth is not a lot. The professor posts a lot of good optional reading material so students can explore interesting topics in greater detail on their own. Because the course takes on a more overview type of approach, the content is more breadth over depth, so rigorous mathematical discussion and intuition of various ML approaches is lacking, and left as an exercise to the student. About the teaching style, the professor will just come, deliver the content planned and leave without much interaction, so the students have to be proactive and ask doubts wherever required by interrupting the professor (he doesn’t mind at all)
Who can take this course?: I took this course in my 5th semester. This course can also be taken by sophies in the third semester, after which you can explore ML in a more rigorous manner through courses from the CS and IE departments, and through projects under professors.