This course is good. It introduces the concepts regarding recurrent models. I specially like the attention model videos. In general, the exercises are well written. However, I'm not very familiar with Keras and working on the Keras code really takes my time even I'm quite experienced with Tensorflow.

Professor Ng did a good job explaining sequence model and I finally understand the basic theories. However, there is room to improve especially on the Keras library part. I hope you can add some simple tutorial for that. Also, I still don't understand how to translate the architecture to Keras code.

This is a very well structured course. I initially started this course almost 10 months ago but got distracted and started to learn sequence models on my own. But, at the end end I resorted to this course again and got my basics cleared out. Thanks to deeplearning.ai team for designing this course!

Great explanations on the videos, and well designed programming exercises. However, the complexity of the programming tasks is not well dimensioned (1h - 1:30 h may be too little). Worse, some of the exercises are not well explained, with misleading information (e.g. about model tensor dimensions).

There was not enough theory as the first three courses. Some explanations were superficial and difficult to understand. Maybe drawing the shape of the inputs, outputs, and other matrices in lectures would help to better visualize what's going on inside the networks. Overall, it was a great course.

It's a great course. The information and knowledge that you get about Sequence Models is fantastic as a primer. Andrew is an amazing teacher throughout the entire specialization altough I found the content of the videos in the Sequence Models slightly more rushed than the previous 4 courses...

Poderia dar um pouco mais de instruções na hora de usar as funções do Keras. Ficou um pouco confuso.

No assignment da semana 3 (Machine translation). Tem um output que não precisa estar exatamente igual mas o curso não fala isso e acabei gastando bastante tempo nisso, só vi depois no fórum.

Though the lessons are interesting and mathematically demanding, I felt that it was a good time spent learning these concepts. Overall I feel that the 3 weeks could be split into 4 weeks and learning could have been much smoother by adding some more lessons to address the contents of Week 2.

Nice content, but the assignments are too easy and only demonstrate the pipeline instead of providing hand-on experience in picking the network and training with GPU. Also, there are some grader problems which has wasted my time to make my code pass the grader even if the answer is correct.

The homework grading methods need improvement. I got the right model and get the right results, but still have to spend tons of time to make the submission pass the grading system. It is a waste of time for future learning. You may want to train a DL model to solve this grading problem :).

Obviously, The professor and TAs have put a lot of time for preparation of this course, and I really appreciate it. However, the hws of the course is too much focused on language translation. They could put another examples, say business data, to represent other applications of RNN/LSTMs.

The course should provide more instruction on the Keras and Tensorflow, since the notebook is largely dependent on the knowledge of these frameworks. Moreover, the logic of programming is not so well-organized: I personally prefer to have my own logic instead of modules got implemented :)

Really interesting course with overview about sequence models and what can you do with them. Lectures from prof. NG are amazing as usual. The only thing I was missing was maybe more tutorials on Keras LSTM usage. The exercises on LSTM were quite confusing, especially using shared layers.

Very Important !!!

Hi,

We do need more programming assignments in order to master the material. We joined Andrew's courses to master (not just get introduced to) the materials, because Andrew and the rest of the team is awesome.

Thank you very very much for all your time and consideration

I am overwhelmed by too much material. Additionally Tensorflow and Keras syntax is not very elegant or coherent as they are such high-level languages. I learnt a lot at a high-level overview in this course, but my fundamental understanding was consolidated in the previous 4 courses.

Finally, the last course was completed. For me, this course is very difficult, because the content of the course is somewhat obscure and difficult to understand. But I learned some basic knowledge about Natural Language Processing and Speech Recognition through this course. Thanks!

This is my third course in Deep Learning, the contents and pace of learning are great, they provide a good level of understanding in the subject. The notebooks have bugs and I wasted a lot of times making them work, thus they could be improved to use that time actually learning.

The concepts presented in this course were advanced enough. Yet, the assignments did not require much effort and thinking, which in my opinion is hurting the learning process. If students do not struggle enough with the course, they tend to forget the material they have learned.

Very interesting courses. I take this as a basis for future applications. I only regret that the exercises are too guided. I can't pretend to be able to accomplish a project in machine learning :-(

I would recommend also to note all the references to the papers, they are helpful.

Great content, assignments are fun and reasonably instructive (although they contain the occasional error and the video editing for the lecture content seems a bit rushed at times). I would recommend this course as an introduction to recurrent neural networks and related ideas.

Very informative and well taught course on sequence models. The amount of content and pacing was just right as not to be overwhelmingly complicated. There are a few bugs here and there in the programming exercise which can lead to a lot of headaches but overall a good course!

Week 3: quite a complected network was used for trigger word detection; however, it is not clear why exactly this architecture was used; specific order of dropout, batchnorm and GRU seems to be a pure magic; at least, a few words why this combination is picked are needed.

Course content is excellent, I would have given 5 stars, if the Programming assignments wouldn't have bugs. Fortunately people in forum help out with solving issues with assignments. I believe it's due to the short time frame the course is online and bugs get corrected.

While this was the most relevant course for me, I missed how it was focused on "helper functions" instead of core RNN concepts. While I feel like I understand concepts like the Bleu score, I would definitely need to spend more time to fully grasp the RNN architecture.

Would have been nice to get more extensive training in Keras en Tensorflow because programming excercies were somewhat too pre-compiled at times or other times difficult to code because of scarse knowledge of these packets. Otherwise great lecture material as usual