Georgia Tech CS 7641: Machine Learning

I took CS 7641 in Spring 2026 alongside Time Series Analysis, which I came to regret. It is by far the hardest course I have taken in the OMSCS program up to this point and deserves your full attention. If you can, take it on its own. It is also a required course for the Machine Learning specialization, so if that is the track you are on you will be taking it regardless.

Prerequisites

Beyond general computer science knowledge, it is hard to pinpoint a strict prerequisite. You will need to be comfortable coding in Python, but much of the machine learning knowledge you need is built through the course itself. I would not recommend taking this as your first OMSCS course. Coming in with some familiarity with machine learning, even from self-study, will make the early weeks much more manageable.

Curriculum

The course is organized into four units, each culminating in a quiz and a major report:

• Supervised Learning — decision trees, regression and classification, neural networks, instance-based learning, ensemble methods, kernel methods and SVMs, computational learning theory, and VC dimensions
• Optimization and Uncertainty in Learning — randomized optimization, modern optimizers like AdamW, information theory, Bayesian learning, and Bayesian inference
• Unsupervised Learning — clustering, feature selection, feature transformation, and manifold learning including PCA, ICA, and t-SNE
• Reinforcement Learning — Markov decision processes, core RL algorithms, game theory, and modern RL variants

Textbook

The course uses Machine Learning by Tom Mitchell as its primary text. It is an older but foundational book that aligns closely with the lecture material and is freely available online. The recorded lectures follow it closely week to week so it is worth reading alongside them.

Reports

The reports are the defining challenge of this course, accounting for 48% of your grade. There are four of them, one per unit, and they are entirely open-ended. No boilerplate code is provided and there is no rubric. You are expected to run experiments, analyze your results, and write up your findings in LaTeX through Overleaf, with a maximum of 8 pages per report. The course treats communication and analysis as core skills, which separates it from most ML courses. Each report took 40 or more hours to complete, and on weeks where I was fully focused on this course I was putting in 30 or more hours.

Start every report as early as possible. If you do not start early you will not have a good time finishing.

After each report is graded you have one week to revise your submission and respond to the feedback. If done well you can recover half of the missed points, which makes it worth taking the feedback seriously.

AI Policy

One thing that stands out about this course is its stance on AI tools. Using AI is allowed and in some cases encouraged, as long as you disclose it. You cannot use it to synthesize your analysis or generate your ideas, since that is the core of what the reports are grading. But it is a legitimate tool for helping with code, which lets you focus on the reasoning and analysis. If you are used to courses that ban AI entirely this policy will feel refreshingly practical.

Quizzes

Each unit has a quiz you can complete on your own time within a window. You get four attempts and can see which questions you got wrong after each one, making it easy to learn from your mistakes and improve. If you stay on top of the material the quizzes are very manageable.

Final Exam

The final exam is hard and covers everything. It is closed book and closed note, which is a different challenge from the open-ended reports. Staying engaged with the material throughout the semester is the best preparation. Cramming will not be enough given how broad the curriculum is.

Grading

The grading scale was one of the fairer aspects of the course. The cutoff for an A was 83, which was the class average. Extra credit is given out generously through things like contributing to the discussion board and completing an optional problem set before the final.

Overall

CS 7641 is a rigorous and rewarding course. It covers a lot of ground in machine learning and the reports are what make it stick. Because there is no rubric and no scaffolding, you are forced to actually implement the algorithms, run experiments, and reason over your results. You cannot get through the reports without genuinely understanding what you are doing, which means you come out with real knowledge rather than surface-level familiarity. Start every report early and do not underestimate how much time it requires. I would not recommend taking it alongside another demanding course.