ECBM E4040

2 assignments , 2 exams, 1 final project

One of the worst classes I have ever taken and reading the other review I kind of feel like they are mostly fake. It is supposed to focus on the theory behind deep learning, but in reality the content is covered extremely superficially. The slides are almost entirely copied from another professor at the University of Toronto: literally the only difference is the name on the first slide. The whole course is essentially a copy of the Toronto one, except that all the practical tutorials (which are included in the original course) have been removed. Instead, they are replaced with 15-minute TA presentations that add absolutely nothing of value. The slides themselves do not go into detail about anything and just show a couple of formulas or pictures. If you have a good math background and really want to understand the concepts, how and why they work deeply, this is not the course for you. Ed is not very active, and responses often take days to arrive, which is especially frustrating close to deadlines. Deadlines themselves are unclear: sometimes the Gradescope submission only opens a few days/hours before the actual deadline. In both assignments, the problem statements (also copied from the Toronto course) were released late compared to what was originally communicated. The professor is not engaging at all. The exams are unfair and poorly aligned with what is actually taught in cla…

2 assignments - 20 % each = 40 % final project (team) - 20% 2 quizzes (exams) - 20 % each = 40 %

Professor Zemel is a nice and good teacher. However, this class is very disorganized, in fact the most disorganized class in the cs department. Ed discussion questions are never answered, submission on gradescope/ canvas open 2 days before the deadline. Only slides that doesn't help as much, no utilization of notes or board & chalk. TA's hold office hours only when close to an exam or hw is released. No material get posted on time. If you know the material you will be okay, if you are not sure and want to learn it - I would recommend taking it with other professor. However to offset for the disorganization, he curves the class to anywhere between A- - B+. I do not think it helps as much since most students in the class are MS students who skew the grade higher. TA's are meant to hold office hours every week and that is not the case. Exam is more concept and numeric. The hws are mix of conceptual and coding homework. Hw's are easy, exam is medium. Overall: You are going to be learning the material by yourself.

As mentioned above, 2 assignments and 2 quizzes. They take time to grade them normal amount of time as other classes. Since the assignments are sort of heavily loaded, they take a bit of time, which is fair. One thing that will help you a lot is to review all the lectures consistently (at the end of every week if possible) and ask questions about anything confusing during the lecture. The professor encourages us to interact, and he answers all the questions as always. A great class overall! (Regardless of what grade I might be getting from the midterm, the quality of this class is very satisfactory.)

It technically is called either COMS W4776. He is the best COMS professor I have had so far at Columbia. He teaches very clearly in his lectures, and his students' tutorials help a lot. There are only 2 big assignments and 2 quizzes (midterm and final). As long as you have good foundational knowledge about ML in general, you will do great. The midterm was pretty relatable to the lectures. Just make sure to review everything, like writing down the key concepts. You will be fine. No coding on the quizzes, but there is a large portion of questions in the assignments. Only one thing I want to mention is that TAs do not have any set schedules, which is not that helpful. They only announce their schedule before the exam to help us with the preparation of the exam! Other than that, the Professor always consistently holds his office hours which is very helpful.