Much more challenging than the other courses in the DL specialisation. It forced me to delve a little deeper into the topic in order to overcome obstacles in the assignments. Content-wise, it's a great introduction to DL for NLP. Professor Ng's explanations are perfect.

Admittedly, compared to the other courses, this one is "messier". Spelling mistakes, some contradictory instructions, and a somewhat broken notebook for the last assignment. It felt rushed and I'm surprised that a lot of the errors haven't been fixed yet.

The course itself is cutting-edge, so a 5-star for this.

But the following amount to a -1 star:

1 Too sloppy, lots of typos.

2 Wrong answers wrong expected values in the notebook.

3 Grading server sometimes runs slow.

4 Saving the notebook fails quite often.

5 Too much is done for the learners, while you could've make the programming assignments more challenging.

6 Deep learning itself has too much black magic and inexplanability in it.

I'm quite sure that harsher comments and a few 2-star or 3-star will be among the reviews.

In general, about all the specialization, I think that some of the programming assigments could be more didactive to understand the concepts of the courses and not to find what the code is doing in that specific task. For all the others aspects of the course I think it's perfect for a litlle bit more than an introduction.

For the last course, I feel that some concepts were explained very fast and for some of these i took me a lot of time to understand what I was doing.

Congratulations for all the good work you made.

The lectures were very helpful - Dr. Ng has a gift for providing context and intuition. The programming assignments were not as instructive as I would have hoped since they mostly involved "fill in the blanks" and matrix math operations. I would have liked more "ground up" development that forced me to learn how to create solutions on my own vs be dependent upon scaffolding and examples. I know that's hard from an automated grading perspective, but perhaps one keystone assignment per class makes sense.

It's a good intro to RNNs (LSTMS and GRU). Very interesting use cases for RNNs. I feel that there could have been more room to try more programming exercises for different use cases & RNN architectures. Be aware that Keras is very sensitive to changes and you will find yourself reloading the jupyter kernels repeateadly when you get stuck. This is not a problem of the course itself but it is something that could end up wasting a lot of your time chasing problems when your code actually should work.

The lecture videos are great. Andrew is a great teacher in teaching concepts at the most intuitive level. By the time I finished the course, I had sufficient knowledge to be able to read and understand the paper.

One shortcoming of the course is that - the applied coding skills you gain from the course are quite limited, as the quiz and the assignments are a bit too simple. I wish there could be an optional advanced version of quizzes and assignments that are more thought-provoking.

Another great course by Andrew Ng! This course is part of the CS230 class currently being taught at Stanford University. Only reason for 4 rather than 5 stars is that at this stage (April 2018), there are few knowledgeable mentors and virtually no Instructors present in the Forum. Course provides little introduction to the syntax of Keras, which makes for some problems implementing models. Therefore one might spend a lot of time spinning one's wheels until finding a way forward.

This last course is the most difficult challenge in the entire specialization, I personally have to watch back the video lectures many times. But I appreciate Andrew's efforts in transmitting knowledge in the most possibly understandable way. In the assignment section, I found it necessary to have advanced experience in Python to be able to comprehend. Even if you have completed all the exercises, I believe most of you would only understand about 30-40% what you were doing.

As with all the 5 courses in the Deep Learning Specialization, the video lectures were amazing, thoughtfully designed (and separated) and gave an understandable overview of the content. As for the programming assignements, some lack a clear description of what is to do - that mostly concerns single steps withing a sub-task though. Tensorflow and Keras need a considerable amount of self-study next to the lectures to truly understand what you are doing there.

NLP is definitely my favorite application of DL. This course was fun, although W4's assignment was quite confusing. Maybe it'd be nice to clarify that though the labs are ungraded, they'll help the learner deepen understanding of concepts and to explicitely recommend that they're completed before the assignment, so overall learning process becomes smoother.

But in general the course and the specialization were quite fulfilling, thanks a lot to the staff.

Theoretical knowledge gained during this course is good, for beginners in this area they learn what can be done and how. Only minus is practical assignements which are too straightforward, it's helpfull to have guide what to do, but maybe some assignement which doesn't have too much of a guide could be good (not necessarily the one which is graded but something which makes you to think a little bit more rather than follow the tips and writing Python code).

I enjoyed this course very much. The videos were very informative covering a lot of ground in RNNs. I also enjoyed the assignments which covered both implementation of RNNs from scratch to get a good feel for it, and practical implementations. I was a bit disappointed about NLP section as it brushed over word embedding and left me without much understanding on how they are estimated. I would also like to have seen time-series covered in this course.

I loved all of the courses in the specialization. However, last two (sequence models and convolutional NNs) had in my opinion poor exercises, not well described, or emphasizing the parts which are not that important, omitting at the same time more important topics. For example the last exercise with spectogram was mainly focused on preparing the data rather than explaining/practicing algorithms. All in all, I gave 4/5 which is still very good grade.

I dont understand why notebooks are become unavailable when I am working on it. It says method not allowed and then please login through www.coursera.org. Then I had to run all the cells again. I think this is because of the lack of resources like # of servers available. Other than that, like the content of the programming assignments. Especially the trigger word detection algorithm worked perfect with my own voice, that was satisfying of course.

This course is a great introduction to sequence models and a great way of finishing the specialization. All main areas were covered. It is a good entry point for those who want to keep improving their sequence-models skills.

Keep in mind that Keras is needed (not a basic level). In comparison with the other courses, the exercises here sometimes are more focused on the preprocessing that in the model itself. Nevertheless, more than recommended.

I got an overview of how people use this technology but the whole network architecture and especially dimensioning remains to be somewhat of a black art.

The overview is much better then one could obtain by downloading tools, or reading framework-centric books. It provides also a lot of information through the references to the scientific literature.

It is clear that this is a field still in its infancy but the results are already amazing.

Awesome course. It feels like this one was more rushed than the others in the Specialization. I am a bit concerned whether the "Specialization" has become a "Survey" of the course, and leans heavily on the assignments for teaching..but then, could do more for why some architectures are the way they are. The assignments improved from being a spoon-feeding exercise. That's good. But, on the other side, Keras documentation was confusing.

The 5th course materials were not as solid as the previous 4 courses. Mainly due to heavily use of Keras API and lack of Numpy implementation. There is nothing to complain about indeed as for the same amount of content the 5th courses had covered, implementing on Numpy will take way longer. Maybe adding more explanation on how the Keras API work will help a tons. Overall it is a very good course that I would recommend anyone to enroll.

Excellent series of courses! Loved the lectures and thoroughly enjoyed the exercises! A big thanks to Andrew Ng and all the instructors and mentors. The forums provided useful hints on the couple of occasions I was stuck. While I would have initially suggested providing more info on Keras APIs, in hindsite the additional time spent in searching Keras documentation was useful arriving at better understanding of the infrastructure.

The Literature for RNN's was not motivating enough compared to Convolutional Networks and the previous courses. However, Andrew did concentrate on the important aspects which would help us in building RNN ourselves. I did feel the assignments were not as elaborative and extensive as the CNN's but I understand the idea behind it (being just to make users aware of the skeleton of the model instead of doing everything from scratch).

This one is much better then the previous one - Coursera team definitely made their homework. However some theoretical blank-spaces are still left. For instance - nothing is said about how recursive gates are being updated during BPTT backward pass. For someone who has the some experience and read some other sources that is not a big deal, but for a complete first-timer who pays attention and uses his brain - this is a pure flaw.

Very interesting application of deep learning. Gives a good overview. Assignments are fun, yet it's too easy to complete the assignments without understanding the big picture. I found the "attitude" assignment in week 3 a bit difficult to grasp however. Particularly liked the dinosaur names, emojify, the humour and positivity in the course: mentioning gender/racial bias, encouraging people to do good with their skills. Namaste!

This topic needs to be longer than 4 weeks. The last section - Transformers - seemed too harried, too much information crammed into the last week. The last assignment was nearly incomprehensible to me, I barely made through it, and was very disappointed because the other 4 courses in this specialization were excellent. You need to develop materials on Attention and Transformers better, maybe make it into another course

Very good overall. Andrew Ng explains the material clearly and accessibly.

I'm deducting a star for occasional issues that get picked up by volunteer moderators on forums, who do a great job, but seem not to get corrected by Coursera staff. Also for one or two small inconsistencies in terminology between lectures and programming exercises. However I've seen much worse and more confusing inconsistencies in other courses.

I thought the course and the specialization was great for people who want to get into the details of deep learning. Although I enjoyed learning about all the details, I wish there was a separate course specifically for Keras and TensorFlow. In practice, we will rarely have to implement the models from scratch, and having a better understanding of Keras and TensorFlow would be more helpful in terms of career prospects.