Intro to Deep Learning

Udacity's "Deep Learning" is a 4-lesson data science course built by Google that covers artificial neural networks. The first lesson builds up some machine learning background on classification problems, while lesson 2 discusses the basic machinery of neural networks and deep learning (neural networks with multiple layers.). Lesson 3 covers convolutional networks for image recognition and lesson 4 covers recurrent networks and issues dealing with text data. This course assumes you have intermediate Python programming experience and basic knowledge of machine learning, statistics, linear algebra and calculus.

Each lesson in the course consists of a series of short video lecture segments with occasional comprehension questions and breaks to apply topics discussed in programming assignments. The video quality itself is good and the lecture quality is adequate, but the lecture segments are very brief, with most lasting around a minute or less. The sum total of the video content in the third lesson on convnets is less than 15 minutes. The programming assignments, which use a popular neural network library called TensorFlow, are lacking in instruction and involve either running large chunks of provided code or working on open-ended questions. You likely won't be able to make much progress on the assignments without prior knowledge of machine learning and TensorFlow or doing a lot of extra research outside of the course materials. The programming problems also require significant computing resources; my laptop with 8GB of RAM ran out of memory when running the provided code in the first assignment.

Deep Learning is a shallow course that is akin to reading CliffsNotes instead of a textbook: you'll learn some terminology and be exposed to some interesting concepts but its abbreviated coverage is likely to confuse students who are new to neural networks while leaving more experienced students unsatisfied. This course seems like a rushed attempt to capitalize on the hottest buzzword in the hottest tech industry, which is a shame because it could have been a good course if it took the time to cover the topics in adequate detail.

I give Deep Learning 2 out of 5 stars: Disappointing.

*If you're interested in learning about the topics this course introduces in much more depth, check out the video lectures and course materials for CS231n, a deep learning course focused on image recognition offered by Stanford.

HORRIBLE QUALITY OF INSTRUCTION -- ERRORS, OMISSIONS AND OTHER ISSUES!
-

I thought I’d try the free Google Deep Learning course on Udacity. Although I don’t agree with doing things poorly, I can at least partially understand that they don’t want to spend a too much time and effort on a course being given away. However, it should at least follow good instructional design principles in terms of what is being presented, how its presented, and subsequently how to test and reinforce the materials.

Unfortunately, this is one of the worst eLearning courses I have ever taken; and it definitely keeps me from considering spending any money on other Udacity courses. In fact, it has so many issues that I cannot even justify wasting time learning. From errors and omissions that will teach people inaccurate information, to suboptimal and wrong approaches recommended for basic ML and related problems, as well as quizzes that pretty much illustrate how NOT to try to reinforce learning; I found it impossible to use-- let alone recommend.

While I didn’t expect a $1,000 course for free, since I was reviewing it for others in need of AI/ML/DL instruction, I hoped it would be a lot more professionally done than it was. I would NEVER put something like this out myself, whether it was free or not-- people’s time and abilities are just too precious to waste or partially derail. GIGO.

We spend between $25K-50K per year on instructional materials (e.g. from Lynda.com and UDEMY on the low-end, to personalized instruction on the high end), , and even the cheapest ones are so much better than this. We have helped create online instruction for online and corporate clients for over 30 years, and this level of instruction is like a throw-back to the early days of “multimedia-based” eLearning. It lacks consistency, accuracy, completeness and any real adherence to well-defined, well-accepted instructional standards and practices.

If Udacity cares (and I see that there are plenty of others who have commented negatively on several of the things I observed and noted below), here are some of the observations:

The questions have no contextual definition, are vague in terms of what level or type of answer is being looked for, and mostly serve to disrupt learning flow, rather than reinforce it.
The support and related materials have errors in them-- even things as simple as summation series definition. That is just sloppy and serves to further undermine confidence in the rest of the materials.
There is a lack of consistency inherent to the level of instruction, which jumps from wasting time explaining material that is widely recognized as self-evident, to making assertions that are either unsupported or unsupportable.

The relations between the instructional materials, the quizzes and the reinforcement learning elements are as disjunct as I have ever seen on “professional” instructional materials.
I wanted to finish reviewing this as part of a more in-depth online Web review for an article, but it is just too painful. I may try to have someone else try to get through it.

My only question is whether this is considered a representative level of quality for Udacity Artificial Intelligence instruction, or is just an example of “you get what you pay for”; and since it is free, there is much less care and oversight given to its quality. I can name many dozens of YouTube vlogs that are much better than this in terms of what they teach and how, and they are also virtually free.

I have taking many classes on Udacity whose instruction quality was exceptional, often employing advanced teaching methods (e.g. actively reflecting on exercises in the "How to Use Git and GitHub" class), so I was shocked about the poor quality of the "Deep Learning" class. In short, "Deep Learning" amounts to little else than a laundry list of methods and concepts of Deep Learning without sufficient instructions on understanding and applying these.

"Deep Learning" is rated as a advanced class and has a long list of requirements. But even taking these into consideration, the class itself does not provide sufficient explanations for new concepts introduced here. I understand that it is important to have the skills to obtain additional resources for challenging concepts, but "Deep Learning" is clearly lacking substance as concepts are introduced very superficially in about 5 min before you are expected to implement these methods in assignments. In the end, you likely end up spending hours on additional research using outside resources before you will be able to solve most of the problems stated in the assignments, which is fine, but begs the question what the benefit of taking this Udacity class is.

Skip.

As many have said the course is poorly put together. It was misrepresented as requiring basic python and little math, both untrue (I believe they have revised the requirements now to be more honest). The material is poorly represented, 80% of it is 3rd party resources they link to. Basically you pay $400 for a list of links to external resources, access to forums and projects that are 90% already done for you.

Most of the content and example code is built in a different version of python than the course requires and using different libraries so is of questionable value

It is also NOT a learn at your own pace course but instead a deadline based one which makes it worse.

I have taken other udacity courses that were amazing. This one is a waste of time

This course has fallen so far below my expectations that it's difficult to describe my disappointment. Within the first week, I had to stop the lesson, do the Coursera DL course, and eventually return to Udacity to complete the first section. This is most likely going to be my tactic for finishing the other sections (And TRUST me, I will finish. I paid hundreds of dollars for this and I will at least use the projects in my portfolio.) I do NOT suggest this course to anyone.

I'm a software developer by trade, with sufficient educational background in algebra and calculus; machine learning should be challenging yet exciting, but instead it's just frustrating and almost painful. There is very little video content and the presentation is quite sing-songy, almost like they're giving an oversimplified explanation of something infinitely complex to a child. The text contains many errors, and lots of references to the Keras and Tensorflow documentation. (Not that you shouldn't read the docs, but why pay hundreds of dollars only to have someone tell you to read something that was already free?)

I honestly was beginning to wonder if my intellectual capacity had diminished because the gap between what is taught and what is tested is so great. If you've experienced this too, don't fret. There are courses available that will help you to feel encouraged about your journey into the field of Deep Learning. It's just unfortunate that this isn't one of them.

I pity those that dropped out. To take this course, you need bags of guts. & this is the best instructor ever!!! He's motivated, insightful and he makes it look soo easy. I'm currently taking the course. I hope to complete it soon. I'm half-way, & i'm loving every bit of it. Although I do believe that the people taking this course must have some knowledge of Machine Learning, it's only fun when you go out of your way to understand the materials. Please make use of the IEEE https://ieeexplore.ieee.org/Xplore/home.jsp academic journals if you get a chance. Academic journal will help you fast track your research/learning process.

It's easy for me because I've been researching Machine Learning algorithms for 2 years now, and also completed the ML courses taught by; Prof. Adrew Ng (Stanford University) & Prof. Geoffrey Hinton (University of Toronto).