Introduction

Carnegie Mellon University offers a diverse and comprehensive set of computer science and machine learning courses. In this post, I’ll be documenting my experience with all of the courses I’ve taken at CMU.

Fall 2024 Semester

10-701: Introduction to Machine Learning

• Content: I liked that the content taught some more rudimentary machine learning and covered statistical learning theory. More and more classes are pivoting towards teaching exclusively Deep Learning, but foundational ML methods and theoretical concepts are increasingly rare in modern courses.
• Teaching: Breadth before depth. A million topics are covered, making you good enough to list keywords on the topic but not nearly enough to understand any topic in much depth. Luckily I had already taken Intro Deep Learning (11-785) earlier and a fair bit of the course was review. But for students whose actual introduction was 10-701, I feel sorry for them. For instance, the two lectures on Reinforcement Learning introduced not only the basics but also an overwhelming array of RL frameworks. So for the final exam, students were cramming a bunch of RL algorithms.
• Homework: This course had the worst homeworks I have ever seen at CMU. There were numerous typos that completely changed the answers to some problems. If one didn’t read a Piazza post, half of their answers could be wrong due to this. I also heard numerous complaints about TAs not being helpful, but I don’t place much blame on the TAs as the course in general had no structure and seemed sporadically done.

Rating: ⭐☆☆☆☆

10-708: Probabilistic Graphical Models

• Content: The content of this course is actually interesting to think about. The class provides a solid foundational knowledge base and introduces a lot of core concepts with mathematical rigor. The Generative AI course covers some similar topics, but I think PGMs provides more mathematical rigor with far less stress.
• Homework: The homeworks were actually quite fun to complete. The problems look very intimidating to a student not well-versed in the notation, but this is easily overcome by reading the lecture slides in greater depth. I found the coding homeworks especially fun as you implement in practice all the methods discussed in class.
• Lectures: Sometimes the lectures feel like a bunch of math equations thrown at you. So it’s absolutely necessary to read blog posts and papers on the subjects to have a more foundational understanding.

Rating: ⭐⭐⭐⭐☆

15-359: Probability & Computing: Randomized Algs and Markov Chains

• Teaching: This class is a continuation of 15-259, but with the addition of Richard Peng joining Weina Wang, the class took a whole new spin. So we got the traditional PnC material from Mor’s textbook, but also a bunch of content from Richard. One day, his content overlapped with a problem on 10-708 HW and I was just amazed.
• Small Class Size: Everyone in class was really funny and it gave the class a great atmosphere. The course will likely no longer be offered due to small class sizes, but I was glad I could take it while it lasted.
• Homework: We only needed to do 5 problems out of a list of 8-10 problems usually. This removed a lot of stress from HW grades as it was your top scores for 5 problems in a HW. Another great benefit of the HWs was that Richard added special HW problems that were notably way more difficult than the textbook problems.

Rating: ⭐⭐⭐⭐⭐

36-700: Probability and Mathematical Statistics

• Useful: While not directly practical, it covers essential statistical concepts like scores, risk, and model selection. These topics are needed for more advanced/theoretical machine learning.
• Homework: They weren’t necessarily interesting, but they were well structured and did not provide many headaches.
• Lectures: They are not terrible but they are not great. For this course I found it a lot more effective to just read the notes and the textbook, both of which are masterfully written.
• Review Heavy: As a CMU undergrad, there was way too much review material in the beginning of the course. This is done to cater to various grad students’ backgrounds. Ideally, motivated students could take 36-705 which skips the heavy review in the beginning, but this class is heavily gatekept and may be too difficult for most undergrads.

Rating: ⭐⭐⭐☆☆

Course Reviews

Spring 2024

15-210: Parallel and Sequential Data Structures and Algorithms

• Interesting: The course was not entirely theoretical as I anticipated. The course covered practical computational topics like granularity control and systems-level considerations. The emphasis on developing parallel algorithms, which was a refreshing departure from purely sequential algorithms.
• Fundamental: The topics covered are imperative for foundational algorithms knowledge.
• Coordination: The TAs and professors (Sleator and Acar) were not always on the same page with the TAs, which sometimes caused tests to not meet students’ expectations.

Rating: ⭐⭐⭐⭐☆

15-259: Probability and Computing

• Lectures: The course material is very well prepared and makes learning probability easy. Mor and Weina finalized a textbook, which is the only resource needed to succeed in this course. Lectures were clear and the course was well-run.
• Interesting Applications: The class is more applicable than the 21-325 version, as it doesn’t discuss sigma fields and more. I found the applications to computing quite fascinating.

Rating: ⭐⭐⭐⭐⭐

15-311: Logic and Mechanized Reasoning

• Teaching: Marijn and Jeremy were good instructors who clearly cared about the course.
• Applicability: Encoding problems into SAT/SMT solvers was engaging, and it was exciting to see state-of-the-art solvers in action. Enjoyed learning about software verification.
• Dry: While powerful, SAT solvers weren’t as interesting as expected. A lot of effort was focused on transformations and proofs that I didn’t find particularly interesting.
• Hard?: This was the first class where I felt extremely confused. I cannot attribute blame to the course as many other students seemed to have a better grasp of the material than myself.

Rating: ⭐⭐⭐☆☆

18-095: Getting Started in Electronics: An Experiential Approach

• Refreshing: Learned more than expected while fulfilling a science requirement.
• Projects: A refreshing change from purely math and CS work by physically building circuits. Some really cool projects for the final hack.
• Teaching: Fantastic support from course staff and Prof. Zajdel.

Rating: ⭐⭐⭐⭐☆

21-373: Algebraic Structures (Dropped)

• Content: Enjoyed the first few weeks and the beauty of the abstractions covered. After that, I was overwhelmed by the terminology later on and couldn’t keep up without significant correction.
• Difficult: The test average was around 50%. This was quite normal and not an anomaly when Bukh teaches.

Rating: ⭐⭐☆☆☆

Fall 2023

11-785: Introduction to Deep Learning

• Teamwork: Great course for those who can form a cohesive group and manage the coursework independently.
• Quizzes: Quizzes are one of the most challenging and time consuming components of the course. They offer hackathons on Saturdays, where students could work together to get the quiz answers. The quizzes can easily take up your whole weekend if poorly time-managed.
• Tedious: HW P2s can become tedious, requiring heavy compute resources to run ablations. It’s very common for students to have 5+ Google accounts to run more training ablations.

Rating: ⭐⭐⭐⭐☆

15-213: Introduction to Computer Systems

• Fundamental: It’s impossible to do any real programming without basic systems knowledge.
• TAs: Knowledgeable and enthusiastic TAs.
• Interesting: Topics include cybersecurity, memory management, caching, and networking.

Rating: ⭐⭐⭐⭐⭐

21-355: Real Analysis

• Content: The class became more engaging towards the end of the semester as we delved into calculus topics.
• Homework: Some problem sets, such as the integration problems, were particularly challenging and required advanced arguments.
• Variance in Course Quality: The quality of math courses often varies based on the instructor, unlike CS classes which are more standardized.

Rating: ⭐⭐⭐☆☆

80-101: Dangerous Ideas

• Framework: Enjoyed the course’s approach to structuring arguments. They teach a framework for arguments which logically makes sense.
• Minimal Workload: The course has a minimal workload for a gen-ed.
• Topics: Some topics discussed were overly abstract and difficult to engage with. To be honest, lots of the topics we argued about in class I didn’t care about. As a result, I found myself indifferent to both sides’ arguments.

Rating: ⭐⭐⭐☆☆

Spring 2023

15-150: Principles of Functional Programming

• Interesting: Many basic CS tasks were solved in the framework of Functional Programming. I found this paradigm quite neat.
• Formalization: Learning time complexity with work and span calculations was a highlight. Before this class, the arguments regarding time complexity lacked rigor.

Rating: ⭐⭐⭐⭐☆

15-251: Great Theoretical Ideas of Computer Science

• Homeworks: The problem sets were super interesting.
• Teaching: Prof. Ada and Prof. Gupta were a great team this semester.
• Supplemental Sessions: Bonus lectures were definitely worth attending.
• Writing Sessions: The idea of writing sessions is to force students to understand the HW well enough that they can write the argument for the problem on notice. This is theoretically nice, but I was not a fan of having to attend a HW session at 7PM every Wednesday.

Rating: ⭐⭐⭐⭐⭐

21-266: Vector Calculus for Computer Scientists

• Content: Great for those who want a formal view of Calculus.
• Teaching: Clive Newstead was a great instructor, and the class was well-organized.

Rating: ⭐⭐⭐⭐☆

79-262: Modern Chinese History

• Interesting: Class is worthwhile even if mildly interested in Modern Chinese History.
• Not easy gen-ed: The course has a reasonable workload and a fair bit of reading.
• Teaching: Prof. Weiner was really engaging and did a great job teaching this course.

Rating: ⭐⭐⭐⭐⭐

84-110: Foundations of Political Economy

• Some Interesting Content: Some mildly intriguing political topics discussed in the class. This included concepts like fairness in voting, which was eventually covered as a bonus lecture in 15-251.
• Reccomendation: I recommend taking a standard Micro/Macro Economics course, which has a more developed curriculum.

Rating: ⭐⭐☆☆☆

Fall 2022

07-131: Great Practical Ideas in Computer Science

• Teaching: TAs were super friendly and helpful in labs.
• Content: A great course for anyone new to Vim, Regex, or Git.

Rating: ⭐⭐⭐⭐☆

15-122: Principles of Imperative Computation

• Fundamental: Provides a solid foundation in Data Structures and Algorithms.
• Fun: The leaderboard on Autolab added an element of competition.

Rating: ⭐⭐⭐⭐☆

15-151: Mathematical Foundations for Computer Science

• Organization: Well-organized with good coordination between TAs and Mackey.
• Fair: Fair grading, with performance on tests weighted correctly.
• Homework: Problem sets had many interesting problems to think about.

Rating: ⭐⭐⭐⭐⭐

21-241: Matrices and Linear Transformations

• Class Versions: CS version of the course covers more content and is superior to the non-CS version.
• Project: Final project allowed exploration of an interesting linear algebra topic. I did pagerank which seemed like a popular option, but the other options were quite interesting.
• Review: The first half of the course was mostly review for those who had taken Linear Algebra in high school.

Rating: ⭐⭐⭐⭐☆

21-295: Putnam Seminar

• Rewarding: Rewarding to solve problems.

Rating: ⭐⭐⭐⭐⭐

76-107/76-108: Writing about Data / Writing about Public Problems

• Not Entirely Boring: Normal freshman writing classes, but 76-108 allowed me to write about something related to CMU, which was interesting.

Rating: ⭐⭐⭐⭐☆