Architecture of Machine Learning Systems

I generally liked the course as it provides a good overview of how systems like PyTorch and Tensorflow work behind the scenes. I was not familiar with topics like compilation, operator fusion and rewrites and it was good to hear about them. However I feel like the lecture doesnt really go into detail in any topic. So be aware that this is more a high-level overview. I did have some experience in using PyTorch so I already knew a lot about the rest of the lectures. In the second part of the lecture they talk about the general machine learning / data science method and I feel like this is covered in every basic data science course so it was not really new. All in all I would say that if you dont know anything about this topic it is a good introduction, however if you are only interested in some specific parts I would only look at the videos or slides as they are all available on the courses GitHub page (they dont really use ISIS). I would not recommend taking the course entirely as I found the grading to be very strict compared to how the preparation lecture made it look.

Also you have to do a project to be able to take the exam. You can either work on the system that the professor is working on (called SystemDS) or you can do a standalone ML pipeline project. I did the standalone project and it was quite a lot, but you have the whole semester for it. It is interesting if you want to work on an ML task from start to finish. It was about classifying hearth rate signals. I also talked to another student that did the SystemDS project and they said that it was very messy and did not work in the end. But I mean it is just about passing the project and it doesnt really matter if it works or not.