An unfortunate class. The professor seems like a solid human being, very much cares about her teaching and impact, and I hope the course improves accordingly. That being said, this was a very poor machine learning course. I would recommend avoiding this class until reviews suggest it has improved. Presently, you are much better off taking an nlp class in one of the variety of online offerings.

In general, I have found there to be a pretty common spectrum of experiences in ML courses. I have seen this in online courses, in undergrad courses at UCLA, and similarly in grad courses I’ve audited. On the one hand, you will encounter classes and professors who are extremely math heavy, with heavy usage of notation that is often assumed to be clear. They are rigorous, but in their best form you will leave with a deep understanding and trust that you know where your assumptions are. On the other hand, you have courses that aim to offer an extremely intuitive explanation. These are difficult to do, and still can provide varying amounts of rigor. I feel like Andrew ng is a very good example here. Kao (at UCLA) is also very strong

This course in its current offering felt like it held a middling position, wherein it didn’t provide clear explanations that help you develop intuition about nlp, nor does it provide any amount of rigor in its mathematical explanations.

The slides are littered with inconsistencies in notation, the mathematical explanations are poor and don’t seem well thought out, and there are so many errors on homeworks and exams that you are left not really knowing what assumptions you were supposed to have and which you shouldn’t.

Violet should consider rethinking her slides, and should reconsider what the overall message she is trying to convey at both the lecture and the slide level. I hope it improves because she has the passion for a good offering.

This was probably one of my favorite CS electives that I've taken at UCLA (there aren't many). Sure, this class was a bit disorganized at times, but the topics covered and the willingness of the teaching staff to listen to students made up for those moments.
Most of the reviews from this quarter (Winter 2024) talk about the midterm and mostly use that midterm to paint their opinion of the entire class. While I do agree that the midterm was one of the few disappointments of this course, the curve, adjusted optional grading scheme, and additional bonus points that Professor Peng gave us all made up for their mistakes.
The midterm aside, this course talked about some extremely interesting topics in NLP. It gave a great introduction to NLP and even gave us practical experience with using state-of-the-art models through the course project. The assignments were extremely doable and the teaching team was very helpful in getting the information across. Professor Peng's lectures were also very informative and she highly encouraged class participation during the lectures, which I enjoyed.
Even if you have a slight interest in another application of CS, I would highly recommend this class!

There are things that can be improved, but this is a GREAT class to take! The previous reviews all express frustration about the midterm our quarter, which I understand as it was insanely and unrealistically difficult. However, the teaching team is very open to feedback and consistently encouraged us to share our thoughts. After it became obvious that the whole class felt the midterm was unfair, they adjusted the grading scheme by making a free-response question extra credit, relaxing the grading rubric, AND we all received a raw 5% boost on the midterm since most of the class filled out a mid-quarter feedback form. The mean after these adjustments was about 83.5% which I feel like is very standard. The final was much easier - they almost gave us too much time to do it, and we still received a raw 3% boost on the final since most of the class filled out the final course evaluation.

I agree that there were some disorganized moments as well, but I think the thing that makes a teaching team great is when they are open to feedback, and that most definitely showed with Professor Peng. We even spent the lecture after the midterm as an open forum for people to discuss their thoughts and feelings about the midterm. I think many professors wouldn't try so hard to understand their students, so I really appreciate Professor Peng.

Besides that, the class really was fun! The topics were interesting, and NLP is super hot right now so it was nice to be able to learn about them from people doing research in that space. I felt that Professor Peng and the TAs were really knowledgeable about the topics and they were always willing to answer questions during lecture or discussion to ensure we understood the concepts. I also don't think the workload was too bad at all. We only had 3 homeworks throughout the quarter, each of which could be finished in several hours. The group project did feel a bit disconnected from the rest of everything but I actually found it very interesting as it allowed us to see more of the real-world applications and experiment with LLMs and prompting. Plus, it was graded very leniently so even if you didn't have time to experiment too much, you could still get full score.

I would HIGHLY recommend taking this class. Just look at reviews from previous quarters, as I think the bad reviews from Winter 2024 are mostly just a reflection of students' frustration about the midterm.

Awful! Do not recommend it at all!
Although Peng tries her best to explain the NLP concepts, she mainly fails to explain the technical stuff. She can just explain the theoretical stuff, and all the maths and computational stuff will not be explained at all. You will have some intuition about NLP, but most of the important details are missed in her lectures. It is totally obvious that she does not have CS related bachelors because she mainly focus on the linguistic part rather than the computer science part, both in teaching and in exams! She loves to ask a lot of tricky, pointless questions in a format of true/false or multiple-choice questions, instead of good conceptual questions. This course had the potential to focus on more fundamental stuff instead of so many memorization questions.

The midterm was bullshit. A lot of heavily computational questions in just 1:30 hours. They did not even have a rough estimate of the required time! The exam needed at least 3 hours to be done. Surprisingly, before the exam, in the TA session, TAs confirmed the midterm would be similar to the midterm samples they released, but they were fooling us! It was in a totally different format. Although she curved the midterm, the average turned out to be 79 out of 100.

The final exam was better than the midterm but still an awful selection of questions to measure one's understanding of NLP. These guys really fail to select meaningful questions for the exams! It was mainly memorization, and a lot of bullshit questions were asked. This time the exam did not have any computational complexity, but they gave 3 hours' time for it!!! The only thing I can assume is that they do not have any idea what the solutions for the questions are, otherwise, they could give a proper amount of time for the midterm and final.

The workload was extraordinary. All quarter long, you have to work heavily on theoretical or practical assignments and the project. The project was totally unrelated to the course material, and you had to learn it on your own. It was not even taught by the TA. The exams and assignments were graded in a strict manner!

Peng is a hack. Explanations make no sense. Doesn't even hold an undergraduate CS degree, and frankly it shows in the quality of the lectures that she is missing some foundational knowledge. This is a technical course and often Peng falls short of the standards set at this school for technical explanations. She is basically useless in OH (you will get the same trash from her lectures). The TAs were mega mid to boot.

Additionally, her English language mastery frankly leaves something to be desired. The lack of fluency when it comes to describing a technical subject further adds to the murkiness of her already terrible lectures.

I survived this class using the online Stanford notes for Stanford's equivalent class. Go figure.

The attitude displayed towards the students was even a bit hostile at times; Peng dreamed up a mirage of a sample midterm, and the median on the real thing was in the 60s as a result (the TAs wrote it, Peng clearly did not communicate with them at all).

The course project is also a complete joke. It has nothing to do with the class. The project material is not discussed in the class. It's basically just a many-hours-long logistics exercise that has very little to do with NLP. It's a busywork time dump that will leave you feeling irritated.

I want to emphasize that I am writing with the most good will at heart. I am not trying to create a mob or attack the instructors. I understand they are trying their best.

However, I feel like I am starting to lose trust in this class because of the number of errors and corrections there have been.

For example, Homework 1 grades have been submitted, unsubmitted like 5 times. The Google Cloud project set up in unclear (i.e. not enough storage), and a hot-fix to the project was just released. 

The midterm was computationally heavy (which I have no problem with). However, during discussion the Friday before the midterm, when asked if the midterm was going to be more conceptual or computational, a TA responded "It will be conceptual". 

It is hard to succeed in a class where I cannot trust an answer given by the TA is the correct answer. More and more of my energy is now being spent on logistics and double checking that HW1 scoring was actually correct, fixing cloud setup, etc. vs. actually learning NLP (which I still think Professor Peng teaches well, allbeit her teaching did not match the midterm).

Well, where do I even begin? To describe my level of disappointment would require the invention of a new word. This class has been a whirlwind of catastrophes from day one till now. The level of disorganization here is not just noticeable, it's practically palpable.

The lectures, if you can call them that, offer nothing but the bare minimum in terms of examples and content, all while hovering at an unreachable conceptual altitude. And let's not even get started on the midterm. It demanded such intricate calculations that not a soul managed to finish it, let alone digest half of the questions. Oh, and surprise! The practice midterm, supposedly preparing us for the ordeal, was a misleading mirage conjured by the professor. The real deal, crafted by the TAs, was light-years away from what anyone could have anticipated.

And as if the midterm fiasco wasn't enough, we're drowning in an ocean of homework and projects. Now, I wouldn't mind the workload if it were sensibly constructed. But no, not only were some questions excessively convoluted and time-consuming, but the sheer volume of errors and ambiguities necessitating constant clarification on Piazza was almost laughable.

Ah, Piazza. The virtual arena where confusion reigns supreme. If you're considering enrolling in this course, do yourself a favor: stock up on popcorn and enjoy the spectacle of students and instructors alike floundering in a sea of contradictions and misunderstandings. It's quite the show.

Really disappointed in this offering of the course. Fascinating subject suffering from a major lack of organization on the part of the TAs and professor. Huge problems of vagueness surrounding homework and project, constant bugs/issues with problems. Midterm was an awful assessment that I feel bad for anyone who had to sit down and take. Highly recommend against this class as an elective until there's major changes.

This class was a letdown - confusing slides, tedious homework, unhelpful hw platform, and a lack of understanding from the professor.

Honestly, it was a mess. The lecture slides are all over the place - no logic, just back and forth chaos that makes studying a nightmare.

Asking the prof about the midterm's ridiculous difficulty? Expect passive-aggressive replies that don't help at all. And for a CS class, the amount of calculation on that midterm was just overkill. Overall, felt like this course missed the mark big time.

Great introduction to NLP and its modern day applications! I learned so much from this course alone. I liked the way the professor explained the material, how it was super engaging and super relevant to how NLP is actually used in the real world. It has even made me consider a career in the NLP/ML field.

The core topic of this course IMO is language models. You start from the very basics (n-grams), and the professor explains step-by-step how language models have evolved over the past few years, culminating at the various transformer models (BERT, GPT).

Grading was very fair and lenient. There were plenty of extra credit opportunities on the projects, and she gave everyone 5 % boost on their midterm and final grades for completing a course survey. The exams themselves were very fair; as long as you understand the lectures, you should be all set.

I also do want to mention the course project, since it allows you to actually have hands-on experience with how researchers are using NLP in the real world. It is a fairly involved group project, but you learn a ton and it could pay off in the future if you choose to pursue something NLP related. That being said, the project takes a LOT longer than one might think, as training the transformer models on the GPUs took forever and the VM environment in which we trained the models kept crashing.