One course to the end: nudging you to complete Andrew Ng’s Deep Learning specialization on Coursera

If you do, you will understand why blurry cats are relevant

You have made it this far. Four out of the five courses required to finish the Deep Learning Specialization. If/when Coursera decides to launch the fifth one (launch date being delayed for more than one month now) you are on your way to be part of the first batch of people accomplishing this.

Regardless of where you are on your path, I wish to nudge you into enrolling and finishing the specialization. I don’t have any association with Coursera nor Andrew Ng (although, as happens with so many online courses, after so many hours watching the same person talking to you, seemingly one on one at your home, you do start to feel, that you are somewhat related to them), nor I think that the course, in this first version is a 100% spotless learning experience…

but there are two reasons pressing me to wish almost everyone I know to complete some sort of fundamental learning on deep learning (subdiscipline of the wider area of artificial intelligence):

1st one: it is pervasive, its current and future implications are discussed on a daily basis

2nd: it runs the risk of becoming one of those areas, with massive impact, that only a few understand or have even have some grasp on (and even academics are making new discoveries each each day)

Obviously not everyone has to be a specialist on this, or even have very advanced thoughts on the topic, but in its basic levels it should well be considered a case of general culture, of something that allows people to reason better over the news they hear every day (including writing them in a more informed way).

I watched a debate on national television in Portugal in which the moderator didn’t seem to have much more knowledge about the topic than the spectators watching at home. Think this should change.

Some obstacles may be running almost immediately through your head:

1st: this is rocket science

2nd: I don’t need this for my work on a daily basis

Regarding the first obstacle, well some things will be admitelly hard to grasp, and some hard to remember, but as the course instructors recall at the beginning of the course, if you remember some algebra and calculus, namely how to multiply matrixes and the meaning of derivatives, that should be the math required at least to allow you a first stab at the topic.

Second, this may be true in most of the cases, but I urge you to also remember that you will be able to understand a bit better the news coming about the topic and also contribute to enlarge the pool of people that have a basic understanding of this, that ideally could range from young people in high school — as soon as they have the basics required, to people in the most diverse professions, and the common citizen wanting to be somewhat versed in the topics of the day.

Now that you are somewhat considering whether this could be seen as a hobby, a dance class that you take twice per week, let’s move on to some of the things that could be helpful, when you are trying to make progress.

• Commit: keep some regularity, come rain or shine, to engage with the materials, the video lectures and the assignments. If much space is allowed between the courses, the conductive thread may be lost, requiring going back and repeating previous lectures to recall where you were. (In this first batch there was some interruption from the 3rd to the 4th course, and now from the 4th to the fifth on the Coursera side, but I imagine that on the future all delays on progressing will be on you).
• Keep moving: just like when reading a book, don’t try to understand all imediately at once, make some pauses and go back when required, but try to keep moving — I have discovered that some of the answers to some quizzes were made clearer on the following courses, if I got stuck where I was, I probably would still be there now.
• Don’t aim for perfect knowledge and understanding at first (or at second) — that is tied to the previous point. If you are expecting to have a complete grasp of something beofore moving on, you may see yourself unable to proceed.
• Don’t declare your learning is finished either. If you do complete the fifth course (whenever they decide to launch it), don’t declare your learning is over. Recall the point above, and now that you have the big picture go back to the particular aspects that still cause you trouble (replay the lectures, review the assignments or do some alternative reading and get explanations from other sources that may drive the point better home to you).
• Recall the other ways in which you are learning as well. This is not only an exercise in deep learning, for many it is an exercise in being able and willing to recall concepts learnt a long time ago, for others an opportunity to engage with online learning and tools like the forums, that requires a skill of its own, in order to be able to extract the best results and contributions to understanding.
• Re(learn to ask). Since the ancient philosophers times that asking has been recognized as a tool of learning, and in the case of self-directed online learning, this has become particulary true. Sometimes you will only get what you need to get unstuck from the questions you and others ask and reply to on the forums. It is impossible for instructors to cover for all the hypothetical issues, specially when are running a course for the first time, so yes, it happens that sometimes, not even replaying the courses or rereading the assignments will help. You have to ask, or have a sense of where to look for a similar questions others might have asked. This skill is transferable to other courses online or offline, and more globally, to life.

Maybe you realize at the end of the specialization that this is just the beginning, of something that you want to explore further, maybe you decide that this big (and perhaps still fuzzy picture) is more than enough for your purpose. In any case you will be a bit more equipped to deal with the flurry of news on AI that appear on your news feed everyday.

I have found particularly motivating to see the amount of people that mention being stuck for hours or days in the same exercise, only to come back with a solution and proceed. In some of the traditional classes, some students have this “forced-to-be-here attitude” and in here, you will see sheer energy applied to learning something new.

I have also been genuinely surprised by some findings when completing the exercises. In this one, of the last course I have completed (the fourth one), when I fed a image from the internet to the algorithm, it was capable of identifying a person (the red square) besides the bus (the yellow square), where the human eye (despite the intuition that there should be one, or more persons), wouldn’t be able to.

In a lecture where the intermediate stages to computing the final outputs of a neural network are shown, via the images the the algorithm identifies has having similar patterns at different stages, it seemed as I was navigating into the subconsicous of a machine, seeing images apparently not having any resemblance to each other being grouped together, and wondering at how on earth the machine was able to find some similarity, some pattern between them.

So plenty of space to be surprised here as well.

Having said this, I do hope to see you on the fifth course, or if you are just starting, wishing you a good ride.