Deep Learning

I was so burnt out taking this course. Please do not double up on this one. Somehow got an A at the end, but did not learn as much as I wanted because I was so tired by the end.

Covers an insane amount of content. Grading largely via auto-grader so not as vague as something like ML. However, performance is a portion of your grade (do not take this class if you dont have a GPU), and there is honestly not very much room to really experiment with ideas because everything is so clearly laid out in the autograder. You will know that a lot of things EXIST, but not actually understand them. Regardless, a good survey course for that. You will be forced to learn a lot, just not that deeply imo.

There is a group project. It is graded very easily. Honestly you can probably get an A on this doing a solo project if you needed to. Pick something interesting and don't worry about getting good results. Find team-members early.

Pro Tip: Give up on the quizzes, the ROI studying for them is not there. You should be getting an A on every assignment if you just put in the work. If you put in the hours you WILL do well.

I spent significantly more time on this course than ML (took over the summer) on a per-week basis.

This course taught me a lot about Deep Learning. Some of the lectures could be better and the Meta lectures are mostly awful; however, self teaching through other materials such as StatQuest or Stanford lectures are very useful. Do not bother reading the textbook, it is absurdly in-depth and is not useful. The quizzes can be studied for using the study guides the course staff releases, but even then, some of the questions on the quiz just make you go "what???". The projects are very interesting but can be pretty brutal. Good course.

Me: I had an undergrad in CS. I started in Fall '24. This was my 4th semester. I've taken:

• HCI. High A.
• KBAI. High A.
• AIES. High A.
• RAIT. High B (I was lazy. It's an easier A than KBAI).
• CN. Low A (I was lazy again).
• NLP. Low A (laaazy bones).

I'm a full-time student. I took this alongside NLP.

DL grade: B (86%).

It's a good course. You learn alot. The lectures can be a bit convoluted. Sometimes things aren't explained well or clearly enough, so supplementary videos may be needed.

The assignments are informative. Not too difficult, but running models for the last assignment took time. But over all, the assignments are fine. Start early and you'll be good. There's a report for each assignment, so don't mess that up. For A4, I messed up the report alot (cuz I was lazy and started late but also the guidelines weren't too clear) and that cost me my A oops. But yea, start early and getting 92+ on each assignment should be very doable.

Quizzes are the most annoying part. Some people did well, but I'd procrastinate and leave everything for the last day so didn't do that well. But even those who studied struggled, so I think the quizzes are just hard. They're worth almost 20% though which is annoying. I'd recommend just going over the lectures a few times, asking chatgpt to clarify things if needed when studying. Then just hope for the best and take it. Many people studied for waay too long without significant gains in performance. They get harder as time goes on so try to do well on the early ones. Pay attention to the focus topics post on Ed that comes a week before the due date.

There's a group project. For me it went really well, we got 100. So just get a competent team and you'll be good. The project was pretty easy. Assuming everyone puts the work in.

If you do well on assignments and the project and do decent on quizzes, an A should be pretty easy ngl. Just try hard enough.

No exams.

TL;DR: Assignments and reports can be time consuming but not too hard. Not hard to do really well on them. Quizzes are hard, but do well on assignments and the project and you'll be good. An A is definitely veeeery achievable.

Courses I've taken in order of difficulty: AIES (1/5) < CN (1.5/5) < HCI (2/5) < RAIT (2.5/5) < KBAI (3/5), NLP (3/5) < DL (3.5/5).

I’d like to confirm the other reviewers’ assertions that there seemed for me to be a disconnect in my prerequisite knowledge and the expectations of this class. As an OMSA student, I took ISYE 6740 a year beforehand and got an A, but it wasn’t fresh in my mind, and I wouldn’t say that I entered the deep learning course strong enough in machine learning and analytic modeling to get an A for this more advanced material. However, it was pretty straightforward to get a B. It really comes down to the oddly difficult quizzes for the difference (I ranged from ~40-90% on them) if you don’t ace the assignments.

Pedagogically, this class isn’t so good; the lectures in the first two modules ranged from pretty good to mediocre, and the third module just had a lot of poor lectures (especially the meta ones). They slap a bunch of research papers, textbook reading, and other random resources on, then tell you to figure out how to read them and discern information without much, if any, guidance at all. With the class being so demanding, the expectations for what skills were necessary were just taught vaguely and it was definitely a course where one is expected to do a lot of self learning. It’s either figure it out or ask questions repeatedly on Ed and there’s not much scaffolding, so I found myself using a lot of resources outside of the class to learn. The first two assignments are helpful and interesting, at least, but like the other reviewer stated, they still hadn’t fixed A4’s weird debugging issues as of summer 2025.

The summer felt like a frenetic pace, especially by June, where there was a proposal, quiz, and A2 due the same weekend, so this arrangement definitely is not for the light of heart. If you’re willing to bust your butt (20-30 hr a week, which I definitely did not do), you will get something significant out of this course because the expectations are pretty high, but since I had to teach myself so many things, it left me wondering what I was getting that I couldn’t get from some of the free deep learning resources readily available on the internet.

This class was quite pleasant in summer with a practical pedagogical style that is much better than ML's focus on conforming to a set of guidelines.

The assignments, quizzes and projects are designed to test your understanding (vs. how to write reports).

The class is front loaded, with probably the most tedious work in A2 over summer. This assignment seems to be designed to weed some people out before the drop deadline. If you're completely new to DL and can make it through A2, things should be fine.

The quizzes can be quite tough, but are a relatively small percent of total grade, so make sure to get 100% on the assignments and project, which gives you slack for worser performance on the quizzes.

My average for the quizzes was around 80%, but I'll finish with a high A.

Summer 2025 , which isn't yet an option for the OMSCentral.

I think in summary, this class has one of the best learning outcomes with some of the weirdest pedagogy outside of ML. To get it out of the way, the quizzes are awful - I literally could not figure out how to study for them using the baseline course material. It seems like either outside course material (i.e. the UMich lectures) or office hours are required for getting a good grade on these, but even still they can be on something like 3-5 hours of material and 150+ slides and you can be asked for trivia on one image in one of the slides. There may have been another way to go about it, but I could not figure it how to do well on these even after speaking with the TAs and trying a different strategy for each quiz ; all in all, they felt antagonistic and weird. YMMV, but it added an unbelievable amount of stress an already high workload class.

Additionally, the lectures are good but not great. There's a lot of expectation of picking up information from them, but they don't do a good job of indicating what's important and what's not for me. The lecture styles of ML and RL work very well for me and I know a lot of students hate them, so think is a personal issue and not necessarily an issue with the class.

The assignments are the meat of the class. They're really time consuming, and each require a non-trivial coding section that needs to pass gradescope, a paper review, and a theory problem or problems. If you are not familiar with multi-variable calculus, expect to spend a long time on these sections to really dial in your understanding. As with the majority of classes in the OMSCS, this is where the majority of learning happens, and you'll get a good sense of how the internals of simple NNs work and how to build your own. For perspective, the last project has you build multiple different types of NNs for language translation and I spent close to a week working through just the coding piece, a weekend on the theory section, and about 5-6 hours on the paper review. It's a beast.

Reports are submitted via landscape PPT slide. I found this format deeply annoying, so I just used the JDF instead. This was fine with the TAs and I didn't get docked for it.

The final assignment is a group project. I liked my group, and we came together with something reasonable in the end.

As of right now, I have a ~91.5% in the class, as I have about a 99% average on the assignments and a 55%-ish average on the quizzes. I will get an A in the class if my Final Project is over a 95, which I sure hope it is.

All in all, this was a challenging course severely marred by some weird pedagogical decisions. I'm very happy with my learning outcome, but I am deeply unhappy with the process and wouldn't necessarily recommend the class.

I have a few ML/AI courses under my belt. I would say DL was the most rewarding courses with a relatively high workload.

The most variable time component of this course are the quizzes. they are worth very little in comparison to the projects, but you can spend a lot of time studying for them. Since this is an optional course for most students, I would recommend spending the time to do well on the quizzes. It is time consuming, but there is a wealth of knowledge to tap into that you would not be incentivized to delve into otherwise, and you will leave with a much richer understanding of deep learning if you actually give 100% to the required course material (quizzes, research paper analysis/synthesis, and projects)

I will say that the 2nd half of the course leaves much more to be desired, and I would definitely recommend taking NLP if you feel yourself not understanding Language Models.

Great course, I got a lot out of it, I would definitely recommend.

I ended up with a B as I've got other things going on in my life and I decided that some things are just not worth it.

Review

Lectures: Very useful at the start when there is more math involved, I didn't find the facebook lectures or the lectures for project 3 or 4 very useful and I eventually stopped watching them.

Quizzes: I studied pretty well for the first one and got 70%, then didn't really study for the remaining 3 and did bad on them, this was the area of the course that I decided was not worth my time/energy, so I did poorly and got a B because of them, but I'm fine with that.

Assignments: I really liked the first 2 Assignments, I thought that they were great and I learned a lot. Maybe I started to burn out by project 3/4 but I didn't get as much out of them. Part of the problem was that you can pass the test cases and still not really conceptualize what is happening, but I think that the project on transformers is valuable, just to have some hands on experience with them.

Final Project: I loved the final project, which I think is unusual from what I've read. I proposed to my group a project that would be applicable to my current job, so I ended up getting really into it and reading a bunch of papers and really trying to make it work. It was fun and exciting because I felt like I had something to gain. My group was fine, I did most of the work, which I didn't expect to do. I don't think that group project work is every really spread evenly so that was ok.

Overall I took so much away from this course: I feel confident working with deep learning models and reading research papers to stay on top of the latest developments. I also have become really interested/passionate about this topic.

I spent around 20 hours a week a few times, before assignments were due, some weeks I put in very few hours, but most around 10-12. So maybe an average of 14 hours per week. But, I also completely neglected the quizzes.