<3

<3

N

A

gj

.

l

I

i

Great introduction to using Tensorflow to implement convolutional networks.

I took the Stanford course by Andrew Ng first, so many of the concepts were very familiar - in some cases, the detail was just a little bit shallow - probably to avoid interfering with getting on with implementation - but this course certainly had references outside the course to some more detailed information on topics like how convolutions help identify features or the learning factor.

The jupiter notebooks were great in that you don't need to worry about the environment much - it's already set up - a big worry for me for many of these types of courses. But there were quirks, and a few times I (and some of the other students) could get tripped up for a little while. If you are a developer like I used to be, then troubleshooting and debugging environment/code issues is a small hurdle though.

Kudos to the instructors and those that set up the course - this is otherwise very hard material to teach and set up good "hands on" evaluation, which they did really well, a couple kinks aside.

The course is a great introduction to the use of tensorflow. Keep in mind, this course is a practisioners guide to Deep Learning, so there's not much theory involved, you just get an intuition to how things work. I'd say this is a great start to the specialization, which I feel will probably compliment Andrew Ng's specialization greatly (that one delves into the theory, but the code you make isn't really what one would use in day to day ML projects).

The only thing I didn't like (why I docked a star) where the programming assignments. Too easy, and the autograder barely checks anything. You need to explore thigs by yoursefl and be disciplined, as the programming excercises let you get away with anything. That said, that seems like it only happens in this first course (maybe as a way to encourage people to keep moving forward), and the excercises get better in what I've seen of the next course.

All in all, great course. Mr Moroney is one of the best teachers I've ever see, and communicates his knowledge and pasion with great ease.

This is a great course with very useful lessons that helps the students feel confident about implementing Deep Learning solutions. It is a perfect follow up for Deep Learning Specialization which lays down the theoretical foundations. The instructor is great, and he talks about real world problems (not just Fashon MNIST but non centered, colored and large images) and explains them very clearly.

There is some amount of lack of attention to details in the course which manifest itself specially in the code (typos, code and code comments not agreeing with each other, and entire lessons which are slotted for 10 minutes or more but dont have any action other than pressing the "mark as complete" button, which makes you feel that you are missing something. Also the discussion board isnt as responsive (especially moderators) as the other Deeplearning.ai courses have been in the past.

I like Laurence's teaching style. This isn't the first his course I've taken. It's nice that he has some interesting datasets of his own and some research questions. But I have a couple of suggestions to this course.

First of all tensorflow documentation has a lot (and I mean A LOT) of good tutorials so I'd expect some of them to be included in the course. That what you can expect of Google's developer advocate to do. I really need some help in understanding those tutorials.

The second suggestion - the course is far too basic. That's probably OK but we really need a follow-up course to dig much deeper into tf.data.Datasets, image processing and custom metrics and losses as an example. It's really not enough background to really reproduce results from say AI for medicine courses that you may get from this series.

The course does a good job of teaching basic TF functionality. It doesn't go deep into actually how NNs work which I supposed is fine if you already have that knowledge. The exercises are a little finicky when it comes to grading. My biggest hangup is that the time they calculate for each week is extremely over-estimated. Half the modules are "Readings" which they allot "10 minutes" for and most of these are a single paragraph or just a rehashing of what was just said in the previous video. The exercises also shouldn't take anyone near 3 hours to complete. So, keep in mind that each week's material can be completed in probably 30-45 minutes at most...not 6 hours!

I thoroughly enjoyed the course and programming various CNNs on TensorFlow. However, in certain lectures (especially the ones with the horse/human data sets), the instructor could spend some more time explaining the process of downloading and storing the training and validation images. It took me some effort and quite a lot of Googling to figure out those parts of the code. While that might not be directly related to the task at hand (binary classification) it is, in my opinion, necessary to understand some of these ancillary tasks as well. Perhaps these explanations could be included as optional videos for those who wish to understand these features of TF.

This is an excellent course to get hands-on. Keeping some tasks as repetitive like those of the callback functions help make the person strongly hands-on and remember them. Just the way, every week's programming assignment involved writing the callback function, if there would be other TF functions/methods that the coder gets to implement and override and other TF abstract classes to extend from, that would have been cherry on top!

Drilling down from the bigger picture of model definition to model.fit seemed extremely useful.

And since there are tons of courses on theory of ML and DL, thank god this one just focusses on coding it out.

I like structure and content of this introdactory course. And like the easy and clear way Laurence Moroney told about all this stuff. Particulary, I like clear formulated exercises. During course we got great bulk of working examples in jupiter notebookes, containg full lecture, notes, likns to supporting materials!

What I would improve in course it is the change a litle bit a balance from solving problems to technical implementation. We learn a lot of using CNN for image recognition. However, it would be great to listen in more details about calculating the shape for input and outputs for layers.

Course in concise and to the point, I was hoping to learn more about TensorFlow than Keras. It is a good course to dive into deep learning without much knowledge in data science. The instructor is motivating and explains the concepts fairly well. I want him to improve the explanation of the parameters (e.g steps_per_epoch vs epochs vs iterations) in Keras as the course is quite applied and making these explanations better will significantly improve the course. I hope to learn more TensorFlow in the next courses in the specialization. The examples were really good and explained the concepts well.

Very good material and enjoyed the short videos, sample code, and ease of moving through the materials.

I like how it is broken into 4-weeks and the amount of effort seems appropriate.

I did experience some frustration with the exercise submission process. Seems whenever I clicked on the link to see my submission, more often than not the submission was not stored and I had to rewrite my solution multiple times. In the end, by clicking 'ok' in the submit response box (rather than the link), I had better luck, as it allowed me to return to the notebook and save it.

I loved the instructors and the content. For the first time, I found a course that actually taught me the practical aspects of deep learning in a fun and interactive way. The content was very good and the right level of difficulty, i.e. not too difficult but also reasonably challenging. One thing I would change about the course to make it better would be to have longer instructional videos that go over all the code in more detail. I did not completely understand some sections of code and I think this would have changed if there were more code explanations.

It's a great course! Very well structured, with an amazing amount of jupyter Notebooks (Colab) to work with, in a real hands on approach.

Just one criticism, which is why I didn't classify it as 5 Star: There isn't much of an evaluation. The tests are a bit easy, and it would be good to have at least one extensive assignment (maybe with other datasets...).

It's just that I feel the contents were really good. But if I can just pass the tests easily, I feel it doesn't really count as much of a "quality stamp" (to have passed this course).

Great course, however it was annoying having to "roll the dice" so to speak to get the answers right. Perhaps if it wasn't left to chance to achieve the right grades, it would of been much quicker and easier to get through. I'd say this is also a course for everyone who's had at least some experience in programming. Understanding some deep learning helps, but you definitely need knowledge of how to code in python. If you can code well in Python and are good with math, then this course would be a breeze for you!

Really helpful and easy to understand. I know several programming languages but not Python, so I appreciated the labs that helped me get up to speed with Python.

The only downside is that I am not particularly interested in image classification, which is the main focus of the course. It was interesting to learn about all the convolution techniques but I doubt I will use them in practice. It would have been nice to have examples of AI problems that were not image classifcation, like a regression problem.

Clear explanations and the tasks, which are easy to understand, make it a really enjoyable journey to learn the depts of Machine Learning. One problem I had was following:

The assignments at the end of each week require me to go to a web-lab, to which I unfortunately couldn't connect. There was always an error which lead to the problem, that I couldn't test the code I wrote there. Therefore I had to copy all data-sets on my own working-station und try it out there and hope, it would work on the lab.