UIUC MCS Chicago - Mid Semester Check In

It’s the Friday before spring break so I thought I’d take a moment to write about how my semester has gone so far.

I signed up for three classes this semester, which makes me a full time student, and it’s definitely been busier than the fall was. It’s taken a lot of weekly planning and a bit of discipline to manage with working full time, but now that break is here, I’m feeling confident and can begin to see the light at the end of the tunnel on the semester, although I’m sure April is going to be a busy one.

The classes I’m taking are CS 521 - Machine Learning and Compilers, CS 534 - Advanced Topics in Computer Architecture, and CS 446 - Machine Learning. I’ll discuss each of them briefly.

CS 521

I think this class so far has definitely been the most interesting, and the most challenging. The class is split into two parts, one focused on compilers, specifically for machine learning frameworks, and one focused more on artificial intelligence and how to use machine learning to build models. So far, we’ve gotten through the compiler portion of the class.

Although the compiler topics themselves I find a bit less interesting, they’ve been good to jog my memory on optimization topics from undergrad and have given us some good exposure to interesting topics through the programming assignments.

In the first assignment, I had my first exposure to programming GPUs through CUDA, looking at implementing optimizations for GEMM kernels. The more interesting part was getting to utilize NKI on the Amazon Trainium platform to implement a forward pass of a Convolutional Neural Network. This was really hard. Mostly for two reasons, the first is that NKI does not have great documentation, so it was basically guess and check. The second is that because it’s so new, the large language models are NOT at all helpful in programming on Trainium or with NKI. It was a good learning experience for sure to get back into debugging as one had to in the olden days, but definitely time consuming. I blew through a lot of our allotted AWS credits for the semester on this one, but it was worth it to gain some experience programming on AI accelerators.

In the second programming assignment, we implemented some optimizations within the XLA compiler. Again, was very interesting to get under the hood of a modern machine learning compiler, because it’s easy to forget how most things today are simply abstracted away in modern ML frameworks.

CS 534

This class has probably been the most enjoyable so far. The specific focus of the class is on mobile and edge computing. I really like the format where in class we go through lectures on various hardware topics and outside of class we read research papers for discussions later. There’s some presentations, some quizzes, and some paper reviews which make up the majority of the grades, and in addition, a semester long project of our choice. The MCS in Chicago people (my group) have decided to perform some analysis and benchmarking of networks on chip for various modern computing workloads. Stay tuned, probably will have more to say on this in the coming months.

Most of the early topics in this class focused on memory, really getting into the nitty gritty on how DRAM works, various caching techniques, and how chips have evolved due to various energy constraints. A lot of these hardware topics are fundamental to computing, but usually taken for granted in CS, so I’ve enjoyed getting into them. As of late, we’ve started transitioning a bit more into parallel computing architectures like GPUs, and will end the class focused on modern trends in accelerators and in edge devices, which I’m looking forward to.

CS 446

Sadly, this has been my least enjoyable class so far, but there’s a specific reason for that. This is my only class which is not in person this semester, and is being delivered on Coursera. There’s been some growing pains to work out, lectures and assignments are often delayed, which has been a bit frustrating. Most frustrating is that our midterm exam they couldn’t get out on time, and now I won’t get to take it until after spring break, which I was hoping to have done first.

The class itself, titled Machine Learning, so far has focussed mostly on supervised learning. We’ve learned a lot of the standard models like kNN, perceptron, Naive Bayes, SVM, and some regression, among others. It’s been good to test my math again (which is very rusty lol), but at times it can be a bit dry. But it’s always good to learn the fundamentals, and I’m excited for some of the later topics when we get more into the weeds on neural networks and deep learning.

So yeah, that’s been the semester so far. There’s definitely been a bit more overlap this time around with work, which kind of sucks, but just something we have to deal with. On the whole, I’ve really been enjoying school though, I like it a lot more than my job currently, specifically because of the in-person aspects (where my job is 98% virtual meetings) and also the requirement for critical thinking and learning. It’s much more stimulating lately than work is. Have a solid 8 weeks to go so let’s hope we finish strong! For now, I’m headed out to Colombia for spring break, so until next time, mucho juicio