Very interesting course. Gives a good introduction to pytorch. My only concern is the quality of the quizzes: It is often limited to 2 very simple questions. This does not allow you to validate that you had a good understanding of the said topic.

At the beginning, PyTorch framework seems very hard to understand. At the half of course you begin to have a clear vision of the problems. A negative point is the notebook for every topic. I would suggest one for week with everything inside.

Good, thorough course. Does not hold the student to any kind of standard or accountability and quizzes are ridiculously easy to pass.

The amount of material was surprisingly extensive and the labs were very useful. The tests were not very good. The videos were OK.

The quizzes are way too easy. The videos are OK (read by computer voice except one). The labs, on the other hand, a really nice.

More graded coding assignments would have been better, but content is good!

It is a nice course to get you into Pytorch and with some insightful views of how some ML algorithms work but adding to the most upvoted review, the synth voice dialogue sometimes doesn't make sense, the inflections on the speech are weird at times, it spells things that come from a text based explanation rather than someone speaking (things like spelling "I E for -for example- and C N N for convolutional neural network among many, many others)... sometimes the voice is talking about one thing and something else is highlighted on the video, time mismatch...

Many grammar mistakes, stuff left in the examples and quizes that doesn't make sense... definitely needs a redaction and content check.

This course had many flaws including that at the most basic it was riddled with errors, typos, and formatting issues.

Some more specific feedback is that this course seemed overly preoccupied with explaining math concepts or neural net architecture at a high level and glossing over much of the actual pyTorch specific programming.

The organization of the lectures make no sense, with separate lectures and labs for single class and multiclass versions of various models even though the functions all were built to handle multiple dimensions and so there was really no difference. Additionally because the lectures, lab, and quiz used all the same examples this means we would see the exact material presented over and over with no clear pedagogical reason.

Additionally the course seemed overly preoccupied with OOP to the point of replicating the functionality of several built in pyTorch classes obfuscating the actual material with no clear reason given for why we were creating our own version of extant classes.

Lastly, the quizes almost never asked any questions about pyTorch. Most of them were just the most basic questions about comprehending reading code. Things like "if input = 3 how many inputs are there?" or "which option is used for He initialization" and the options are like "He initialization or Xavier"

I had very high expectations for this course since it was offered by IBM and being taught by someone with excellent credentials. I completed the course material for the first 2 weeks and I found the lectures to me unmotivating, inadequately explained, and very clearly the lecturer read from a script. Important concepts were not explained neither the conceptual deep learning one nor the PyTorch programming ones. They were very briefly explained often with one short sentence. I thought the ungraded labs were very well designed but the lecture quality was so poor, it seemed I was just googling and learning 90% of PyTorch myself. I had expected quality from this course however, I did not get it so I decided not to pay the $50 subscription and canceled the course. I was disappointed since I did spend good 15-20 hours on this course.

This course is a disorganized and unfocused. For example, much of the section on Bernoulli distribution is misleading or completely incorrect. It's also presented without context. Much of this is redundant give the other courses in this certificate program do a much better job of teaching ML concepts. The novelty of this course is about implementation using pytorch and most of the important details about how to use PyTorch and why certain parameters are used are glossed over.

Is this a course about ML and Neural Networks? Is this a course on PyTorch? It does both poorly.

Please see

https://www.coursera.org/mastertrack/instructional-design-illinois

for how to improve.

I am very disappointed with the quality of the course materials. The videos are recorded with what sounds like a text to speech system or a voice over done by a voice actor who does not really understand the subject matter and lacks personality.

It's hard to understand as it all runs at the same pace and there isn't sufficient time given to specific concepts that may take a shorter or a longer time to sink in depending on their complexity. It's just a constant speed monologue without any real feeling or passion in the subject matter.

it is too deep

This course is really good in explaining the concepts and pytorch. Everything was explained in a detailed way, well structured. However, I found the course too segmented. Some lectures, some quizzes, and some labs can be combined. Example for week 1, I think 1.1 (introduction to tensors), 1.2 (1d tensors) and 1.3 (2d tensors) can be combined to single lecture or all 3 lectures be one after another making it appear like it’s together. The 2 labs can be combined into a single notebook. The 2 quizzes can be combined into 1 quiz of maybe 5 or more questions. Similarly, 1.4 (Simple Datasets) and 1.5 (Datasets) can be combined, and so on. I also think that the honours content about batch normalization should be included as part of normal contents. Maybe more advanced concepts can be put up as honours contents.

even with no mandatory peer graded assignment, for me it was the hardest course to learn in "IBM AI Engineering". So many topics and so many codes to check for each week. i liked it. i believe i will revisit the materials in the future.

Great introduction to deep learning with pytorch. It would help if the notebooks in the labs take shorter to run so that the students can experiment with the code and the models.

this is no doubt THE BEST and the most well thought pytorch and deep learning course so far .

I have learned good skills for deep neural networks

best introduction course on the subject.

Excellent! Just what I needed.

Amazing course!

awesome course

The course is good and has a nice mixture of theory and practice, which is essential for mastering complex concepts. However, I do have a few observations about the course quality:

- Several of the slides in the presentations and even the labs have a lot of grammar mistakes.

- The theory is often rushed in the lectures. The course would greatly benefit from a more careful analysis of the maths behind each concept.

-In its effort to make the concepts easier to grasp, the lectures keep using coloured boxes to replace mathematical terms. I found that to be more confusing, they use far too many colours and are too liberal with their use.

-Lastly, the labs completely broke down in the second half of the course. My understanding from the course staff is that an upgrade was made on the backend which did not go well and thus caused those issues. They should have several backup plans for those occurrences, starting with having the labs available for download so that the students can do them offline.

Overall I'm happy with the course and would cautiously recommend it, given the above shortcomings.

The course was fantastic for someone like me. I already knew all the math, and the course gave deep exposure to the needed Python routines and classes. The labs really help cement the knowledge.

Only drawback is that it went a bit too slow for me (NN with one input, NN with two inputs, NN with one output, NN with two outputs, etc.), but others might disagree.

I'm giving it a four because there were so many typos and mistakes (i.e. the gradient is perpendicular to countour lines, not parallel), lots of mispellings and wrong data on the slides and the speaker sounded like a computer (he pronounced the variable idx as "one-dx" - huh? I understand that there's going to be mistakes, but this is an one online course made for many people, and you'd expect that kind of stuff to be corrected over time since it is being repeatedly delivered.

But - it was a great course and I highly recommend taking it.

Here is a list of pros and cons:

Pros: great notebooks and many examples

Cons: the videos are a bit "cheap" (typos and artificial voice) and often miss the intuitions ("To do that, we code like this"). A bit light on the maths. Quizzes are too easy to validate (people may validate with a superficial understanding of what is going on).

Summary: The value of this class resides in the notebooks and in the time your are willing to invest in them.

I think it's a good course if you're coming in with the notion of deep learning pretty much clear and are more interested in learning the PyTorch syntax. I'm not sure how useful the course would be in terms of learning ML or DL from scratch. In particular the conceptual slides could be better.

The notebooks are well-prepared. Even though occasional bugs can be found, they aren't much to worry about.