I completed the course using Keras and I found one issue that the videos were available for PyTorch only and not for Keras in Week 2,3,4 making it difficult to understand for someone who is new to it. Otherwise everything is good. Great course. Thank You.

Very well.

If your goal is to learn, I recommend having a look at both PyTorch and Keras options and taking notes since some new functions are mentioned.

I trained the models offline on my paptop with GPU and it only took a couple of minutes :)

Really good course, although I had some problems with IBM Watson Studio (Tokyo Region) in latency and constant restarting. I think it was a good introduction for people to get into the industry, many covered.

Course Project was more challenging than others in courses in the Certificate series but that was great! Option to complete project in either PyTorch or Tensorflow/Keras was appreciated.

The Keras part of the course is more attractive just because its final assignment is much better structured than that of PyTorch.

I like the flexibility to pick our framework for the project i wish the kers one were a little bit more challenging

It is a great task but the tools delivered very complicated. It is sometimes better to use upfront your own tools.

I have completed this course but did not get the badge for it. Is there any way to access it?

Thank a lot for creating this course. It really useful and practical for me.

Buen curso, implementando todo lo que se vio en la especializacion

Nice idea, however it could be a little bit more elaborate.

The Keras module was awful to complete

gud

If I hadn't taken Dr Ng's courses first I don't think I would have grasped the concepts, most of the really difficult topics were not covered in any depth and there were leaps between topics that I think could have been covered. I am sure the instructors knew what they were talking about but they tended to gloss over things that ended up being important. I am not really sure about the value of the certificate, one of the students I reviewed had obviously just copied and pasted his responses directly from the rubric. You could tell by the text formatting where the rubric had some text bolded that would not have been bolded by the code. I mentioned it but I think he probably got credit for the work. I don't know if it was my place to not give him the credit per the rubric.

1. The whole specialization is unchallenging. There are no real practical assignments, everything is solved for you. At most you're asked to change parameters in the already finished code.

2. Slides and notebooks have lots of typos.

3. The peer-reviewed assignment for AI Capstone Project has misleading instructions for grading which haven't been solved for over 3 years (https://www.coursera.org/learn/ai-deep-learning-capstone/discussions/weeks/4/threads/-4y66yflEeuDNgpyJIMFiQ).

Overall, most videos in specialization are a good introductory overview of the topics but you'll not gain any valuable practical experience with deep learning.

This course was riddled with operational flaws regarding the image data, and how it operated in the IBM framework. At one point I was not able to run the labs with either PyTorch or Keras versions, and eventually just downloaded the notebooks and ran them in Google Colab to complete the specialization.

The data from the course is not well prepared, some invalid pictures in the data. And also sometimes the IBM platform can not run the training well, loss connection and need several hours of time for training the model...

there are some issues incl.

- IBM lite version crash (So I used my local GPU environment) - Want a more challenging project with friendly provided reference and help

The docker environment by IBM is horrible. I just got to finish my course running all the notebooks locally (except for those at the Watson environment)

Not well-prepared materials in Keras, especially in Week 3 (model-training) which took more than 3 hours to training and even not successfully.

some of the modele and code is outdated.

good course

The content of the course is very interesting and highly informative, however there is a critical flaw in this course (at least for the keras library side of things), the problem is that IBM Cognitive Labs, the intended environment for the assignments, is incapable of running the later labs (week 3 + final) and will crash after 30+ minutes of waiting, this being due to the instructors having us use a relatively large database of images (~250 mb). Jupyter Notebooks on IBM Cognitive Lab struggles to just unzip the dataset (which is downloaded as a zip), not to even mention fitting the models to the data, which I found to be impossible to do with IBM Cognitive labs (for both week 3 and the final assignment). Ultimately I ended up having set up a jupyter lab environment on my own laptop, the problem is even then it took about 14 hours to fit the data to the models (in total, both week 3 and final assignment).

TL;DR the instructors have us using a pointlessly large dataset images which serves more to test our patience than our ability to create deep learning models.

Though this is the most interactive course of the certificate, the nature of the web host for training machine learning models likely prohibits many from actually being able to finish this course. I had to use Google Colab to train the models, as the IBM hosted site would take >10hours to train — in reality it would log you off before the model ever finished training. Due to this, most of the time was not spent on the projects, rather, they were spent re-configuring things to be run with a separate host. I've got degrees in CS so this wasn't a concern, but if you do not have prior experience in CS I would recommend taking a different certificate until this is fixed.

Very disappointing. The instructions are unclear in the assignments and it got frustrating choosing which platform to use to speed up the process and to bypass notebook errors. This was the least challenging and least interesting Capstone project I have done with IBM.