In-depth deeplearning.ai Specialization Review

So you're interested in learning deep learning and you've checked out the deeplearning.ai specialization, but you want to know a little bit more. Well, you've come to the right video. What you see here is, I want to learn deep learning. So that was me, and it could be you as well. But what, what I think you should start with, and this is what this video is going to go through, it's going to go through the why, what, when, how, and where of the deeplearning.ai specialization. But What we're going to start with is why. So, I want to learn deep learning because— and I think, I think you should start here. Figure out a reason why you want to begin. Why? Why should you start with why? Because when it gets hard, that way you can come back to your reason of why you started. Maybe it's to get a new job, maybe it's to build something amazing, maybe it's just because you love new technologies. But whatever the reason is, they're all valid. It's important to begin with one. That way you have something to refer to when, when the motivation is not there and you go, "Oh, why am I even doing this course?" You come back to this, you write it down on a piece of paper or a whiteboard so you can see it all the time, and you go, "Oh, I remember why." So to answer the rest of the questions like what, where, how, and when, Learning's a journey, right? So I figured, why don't we go on a little trip? We'll answer them there. Alright, so you've got your why. What do you actually study? The course is split into 5 parts, and so in Part 1, it's kind of like an overview of deep learning. You get to know the ins and outs of what makes up a neural network. If you're not sure what a neural network is, that's okay, it gets explained in the course. And you'll also You'll also learn how to build your own neural network using pure Python and NumPy, which is Python's numerical library, aka it's used for doing math calculations, because that's in essence what, what deep learning is, right? It's, it's taking one piece of information and using numerical programming to figure out some patterns and relationships in that piece of information. The first programming assignment at the end of Part 1, you'll take that knowledge that you've learned through the lectures and and then you'll put it into practice. You'll build your own neural network using, yeah, using NumPy as we talked before. After you pass Part 1, you go on to Part 2, which is hyperparameter tuning, optimization, and regularization. Now, what do all these terms mean? Well, just think about it as little deep learning tidbits that you can use to improve the performance of your model. So it's just kind of like pushing dials on a car. Like, say you wanted to get a song to play. If push the wrong dials, the song won't play. But if you push the right ones— So Part 2, you'll finish off with a programming assignment which will introduce you to TensorFlow, which is one of the most popular deep learning libraries. So you'll build off what you learned in Part 1 with pure Python and NumPy, and you'll start to use a deep learning framework which helps to shorten the amount of code you actually have to write. So what that means is, is someone else has written a bunch of functions that allow you to write smaller programs that do essentially the same thing as what a quite large program would do. That's where you see those articles that say, uh, state-of-the-art performance in 11 lines of code. It's because really those 11 lines of code are accessing something with thousands of lines of code, such as TensorFlow. And of course, once you pass the quizzes and programming assignments in Part 2, then you can move on to Part 3. Well, actually, you can, you can do them in any order, but I think they work really well one after the other. I have to wind down the window to get out of the car because the door handle's broken. Really need to get that door handle fixed. So what Part 3 takes you through is how to structure a machine learning project. Now this is one thing that really separates the deeplearning.ai specialization from other courses. Can you hear that? It doesn't sound very good. That's really disappointing because this was such a pretty clip. Look at it. You go through this and I walk along the waterfront for a bit. Anyway, what did I actually say in this one? Because I went through it all and the audio is terrible. I got to get a mic. I'll fix that for future videos. So part 3 is what really stands out for the deeplearning.ai specialization, or one of the parts that stands out. I mean, it's all a great course, but this one especially, because Andrew goes through a list of things or goes over some things he's learned structuring machine learning projects. And what does that mean? Well, okay, you know how to build a neural network, you know how to build a machine learning model, but how do you actually weave that into a project? And that's what Andrew shares. It's his learns of actually you putting these techniques into projects? What do you do when it goes wrong? How do you, how do you set your model up so it's, it's robust over time? Like these things, or how do you improve it, right? Like these things you can really only learn after some trial and error, and that's what Andrew distills through this 2-week part of the specialisation. And so at the end you'll have a programming assignment that you'll go through a couple of case studies, and you'll kind of be like a, what's it called, a flight simulator. That's what they call it, a machine learning flight simulator. One is, is you're a, you're a mayor of a city and you've got to decide what type of machine learning project or products or technologies you should use to figure out how many birds are coming into the city. Hint, it's a computer vision one. And the next one is to do with autonomous driving. You get to think about should you use fully automated or fully end-to-end deep learning in an autonomous driving situation. That's essentially what I said in this really pretty section, but you can't really hear it. Back to the main video. Alright, so you've practiced building your own neural network from scratch, you've been introduced to TensorFlow, and now you, you've sort of got an understanding of how machine learning projects work. By the time you get to Course 4, you'll be really getting your feet wet. Because Course 4 is all about computer vision, more specifically convolutional neural networks. And when it comes to computer vision, convolutional neural networks— they're a bit of a mouthful, let's just call them CNNs— are what is really driving the field forward at the moment. And for the programming assignment, well, actually, first of all, you'll be introduced to convolutional neural networks, the ins and outs of them, what it takes to build your very own. And you'll also be introduced to Keras, which is another deep learning library built on top of TensorFlow, and TensorFlow is built on top of Python. And so Keras helps you to really get some beautiful neural nets happening as soon as possible so you can test them out. One of the best, my favourite programming assignments for it was implementing the YOLOv2 algorithm. And I don't know, you can probably see this, this little drone I've got flying around with me. The YOLOv2 algorithm is specifically for object detection, like forming those little boxes around people you might have seen before. And so, that helps follow people, follow things around. If you're a self-driving car, it helps you detect what a street sign is, what a light is, what a person is, what should you avoid, what should you not. So if you want to learn about computer vision, this is the course for you. Oh yeah, I forgot to mention You work on a few other projects such as facial recognition. It's not just the object detection. Like, you'll even build a small little program that can detect whether someone's smiling or not. So, the courses aren't just limited to the one project that I'm mentioning. There's a few, there's a few within, but I thought I'd share my favourites because I can only really give direct insights on what I found most interesting. So, you've wrapped up computer vision. Your computer can see now. What if you want it to be able to hear or read? What do you, what do you do for that? That's where part 5 comes in. All this review making is making me hungry. Let's see if we can go get something to eat. I know just the thing we can use. Hey Siri, where's the nearest cafe? So we've talked about CNNs and computer vision, but sequence models help out whenever you've got data over a time series. So imagine for technology such as Siri, you've got sound waves happening over a certain period of time, or other technologies such as Google Translate where you've got words in a series, or even natural language processing or NLP, which is you could do a whole corpus of text such as all the movie reviews on IMDb and try to work out whether they're positive or negative. The main part of all spider-ments, the sequence of them, and more specifically It happened again. I was recording in a cafe and I didn't realize how loud it would be. You can't— yeah, nah, there's a baby in the background. Yeah, nah, so we'll do it from here. That's all right. Where were we? Part 5. Yes, sequence models. So the main part of Part 5 is, is learning to build recurrent neural networks. All right, so you just like the other ones in CNNs and vanilla neural networks, you learn to build a recurrent neural network from scratch first, and then you look into GRUs and LSTMs. Now, GRUs are gated recurrent units. They're just a— just imagine them as one layer of the neural network. And same thing with, with LSTMs, long short-term memory units. There we go. It takes me— it took me a while to get all the all the jargon and whatnot within, within deep learning. So if it takes you a while, don't worry, because there's lots of terms, and it's just really about practice, just getting them, getting them in your head. And once you've covered the basics of recurrent neural networks, you'll then use that knowledge that you've gained to build your own projects. One of the projects is machine translation. So much like Google Translate takes a sequence of words— remember, this course is about sequence models— takes a sequence of words in one language and converts it into a sequence of words in another language. So, they're more than likely using the same technology that you're using, recurrent neural networks, or some version of LSTM, or slightly tweaked, right? That's the beautiful thing about neural networks is that you can, you can chop and change different layers and, and make it better or worse for your specific task. And then probably my favorite project of the whole course, because I love anything to do with with communication was the trigger word detection one and the speech recognition, just because I've been having so much fun with smart speakers and— Oh wait, what's trigger word detection? It's this, it's where you say, "Hey Siri," and it activates, right? So you'll learn how to build a trigger detection model and subsequently lead on to speech recognition because that's how most of these smart speakers work, right? They need a trigger word to begin with. They're not quite smart enough to to just pick up our ambient voice and be able to react from that. They need that little kickstart, like, "Hey Google!" There we go. Thanks! You're welcome. Alright, so you've got your why and you know what you're going to study. Where do you study it? Now, the beautiful thing about this course is that all the video lectures are available, including the interviews. Oh, that's a little extra. There's video interviews at the end of each course for the first half with, with a deep learning superhero like Yann LeCun, Yoshua Bengio— actually, yeah, I think he's in there— Yoshua Bengio and Geoffrey Hinton. If Yoshua Bengio is in there or not in there, I apologize for getting your hopes up. But you get these interviews at the end of each of the courses, right? And it's Andrew Ng talking about the future state of deep learning with, with some of the best in the field. And all of the video lectures as as well as those interviews are available for free on the deeplearning.ai YouTube channel. But of course, if you sign up to Coursera and do the deeplearning.ai specialization on there, you'll again have access to all the video lectures, all the interviews, the quizzes associated with each of the course, and of course the programming assignments associated with each course. Now what about location? Because it's online, you can study wherever you want. You could study in your own room like I do here. You could go to the library. Shh! Or you could go to the café and study. I had a perfect segue for this café clip. I was even— I even had my iPad there as a show-off to show that you can study on your iPad at a café. Except the audio is not that great. So, now you've got your reason why, you know what you're going to study through the course, You know where you can do it: YouTube, Coursera, library, your room, anywhere you want. When can you do it? You can close this video right now. All the courses are available right, right away. If you just search in the search bar above, deeplearning.ai, you'll find all the lectures. You can jump in and start right now, give it a try, right? Before you even sign up to Coursera, you can try and watch them on YouTube and see if it's for you. But otherwise, if you do sign up to the Coursera cohort or the Coursera specialisation, every couple of weeks they have like an intake of students. So, realistically, you could start every couple of weeks and what that means is you'll be starting with a group of students around the world at the same time. So, if you jump onto the forums, you'll have people who are at the same level as you, asking similar questions to what you are. It's a great way to sort of have a community to to bounce off without exactly being physically around that community. Because that's one of the things about online study is that it can get pretty lonely sometimes, right? Because if you, if you are, if you do choose to study alone in your room, it can be a bit hard if you wanted to reach out and ask someone. So, that's, that's another point. We'll get into that in a sec. Alright, that's, you know, when, when you can do it. How do you do it? If you sign up to it on Coursera, How much does it cost? It's $64 per month, and the allotted time frame that they recommend is about a month per course. But as you heard before, I did it a bit faster than a month per course. So if you do it a bit faster, you might only end up taking 2 months, which means you only pay $64 Australian dollars twice— $128 Australian dollars for the whole course, which is about $47 US dollars. Now there is financial aid available for those who need it, so if reach out and when you sign up and ask Coursera support, you may be eligible for financial aid. But check out, there's some terms and conditions you have to satisfy to be eligible. Alright, I get it. So if you do pay, how's this course structured? How do the lectures play out? The course is broken down into 5 parts, but each part has a week worth of lectures followed by a quiz and an assignment. Each of the lectures is about 5 to 10 minutes in length. Well, actually they can go up to about 20 minutes. But usually I'd watch them on about 1.5 speed. If I really didn't understand something, I would go back to the normal speed, but I got pretty used to watching them at at least 1.5 speed, and that means I could usually complete a week's worth of lectures in about 4 hours of solid study. Alright, I see. So, there's a few different styles of lectures. There's short and longer ones, but I'm up to a programming assignment, what happens if I need help? When you need help, the forums are your friend. You've got— Andrew does an excellent job at explaining some of the things, but more than likely you'll probably come across something you don't quite understand, or there'll be a problem or a challenge in one of the programming assignments where you need a little help. And that's where you head over to the forums. By the time this video comes out, there's probably 100 or maybe 1,000+ people, I'm not sure the exact number, that have gone through it like yourself and have asked questions, and more than likely what you're having a problem with, someone else will have the same problem. So chances are your question has already been answered, but if it's not, don't be afraid to ask, because no matter how insignificant you think your problem is, it's still a problem. In the past, I've been held back from asking out for help because I thought, "Oh, I should know this. This is such an obvious question." but that's held back my learning journey. The only stupid question, or the only stupid thing, is thinking that your question is stupid. There are no stupid questions. Ask for help. Alright, I see, yeah, use the forums, but what happens if I'm really stuck? Like, the forums aren't helping and I'm getting frustrated because I can't do it. Don't forget, when studying gets hard, take a break. Go read something you're interested in. Go do a workout. That's one of my favorite things to do. Take some time for yourself and let the knowledge of what you've just gone over sink into the back of your mind. And you'll probably find that while you're relaxing, the problem you had solves itself in your head. So don't be afraid to take breaks. When it gets tough, walk away. Come back to the problem. All right, Dan. It's one thing to take regular breaks when you hit a roadblock here and there. Like going for a walk or exercising. But how do you actually study it? Well, it's going to be different for everyone. And what's going on here? But I use Trello to track my progress, which, which keeps me up to date. It shows me what I need to do. Oh, hold on. There's a great example of computer vision here. Those cameras in there are scanning my license plate to see if I need to pay. Beautiful! It worked like a treat. That was actually a really good example. So, what the camera was doing was using computer vision, probably a convolutional neural network of some sort, to read my license plate and then compare it to a database to say, "Hey, you need to pay for your parking." That's a lot faster system than having someone there reading my ticket or something like that. But anyway, where was I with Trello? Well, you can organise your board any way you want, but I use it in a relatively simple manner. On the left-hand side, I'll have all the resources that I need for access to the course, and then I'll have a column that's, that's to-do, and then another column that's doing, and then a final column that's done. And what I put in each of them is in to-do, I'll go through the original curriculum of the whole course and make a, make a little Trello block for each one, each lecture or each task that I need to do. Probably for the deep learning specialisation, I'd make one for each course. So, there'd be 1, 2, 3, 4, 5, and then within each of those is a list of the lectures and assessment I need to go through, like, kind of like a checklist, right? And then, as I start part 1, I'll move it into the, the doing column. So, that way, at a quick glance, I go to my Trello board and I know what I need to do, what I'm currently up to, and if I finish part 1, it'll go into the done column. And so I finished that off, that's done. So, I know what I've completed. They're my 2 tips on how I best study is, of course, be concentrated, get into it, do little blocks at a time, take regular breaks if you're getting stuck on something and let your subconscious solve the problem for you, and track your progress. So you know where you're up to and what you have to do next. Whoo! It's been an epic day. What a journey. Kind of like the learning journey you go on if you start the deeplearning.ai specialization. I think to summarize, the main question is, would I do the course again? Yes, I would 100%. I will probably even go back over it again in the near future just to re- re-cement the knowledge in my head. Now, I highly recommend it. It's probably one of the best courses I've ever taken. So if you're interested in getting into deep learning and you've got a few months of Python experience and about a high school math education, or if you've got about a year's experience in another programming language like JavaScript or Java or C or something like that, it's definitely something you should check out. It's, it's one of the best AI courses you'll find on the market. There's a link in the description if you want to go and check it out. I highly encourage you. Coursera offer a 7-day free trial if you just want to go and check it out. And if it's not for you, well, hey, it's not for you. There's plenty of other courses online such as Udacity's AI Nanodegree, Udacity's Deep Learning Nanodegree, and I can only really recommend the courses that I've done. So they're the ones that I've done so far. And if you have any more questions, something I I haven't really answered in the video. It's a beautiful day outside. I just noticed, noticed that. Don't forget, you can send me a message anytime via one of the links in my description. I'll, I'll reply back to you as soon as I can, or if it's a common question, I'll answer it in a future live stream. I encourage all of you to check out the deeplearning.ai specialization, see if it's your thing, and as always, keep learning.