Introduction

0:00 learn the fundamentals of neuron networks from one of the most popular machine learning instructors Dr Ru in this unique handson course you'll learn Core Concepts and manually tweak Network parameters to create a self-driving car while this is phase three of his machine learning series it's fine to start here so let's get started I've been wanting to make this course for six years I got the idea when Grant from three blue and brown posted his video on neural networks the moment he said this one thought experiment that is at once fun and kind of horrifying is to imagine sitting down and setting all of these weights and biases by hand so welcome to my fun and horrifying new course where we'll play with neural networks inside this special playground I created the goal is to teach the car how to drive and we'll do that by manually changing the network parameters using the mouse wheel we'll start with a simple Network that just stops the car from going off Road and gradually increases complexity to teach the car different traffic rules as well there I say you don't need any prerequisites to start this course I

1:10 think the playground and the lessons do a good job explaining the math I use human language and only introduce fancy terms so you know they exist in case you find them elsewhere but if I'm wrong and you still get confused just ask in the comments or on Discord Now by changing these parameters manually will understand exactly what the neural network does this is really different because courses normally teach neural networks in a machine learning context where they're automatically generated from data that makes them work really well but it becomes impossible to tell what they do exactly it's why we call them blackboxes but I want you to understand what the neural network can do before I teach those complex algorithms for generating them and the best way to understand I think is by playing with them like this now this course is good for those starting with AI but also those with some experience who want to understand things better I've been working with machine learning for over 10 years now and some things still surprise me I think it's because training complex models is so easy nowadays just write few lines of code and you're done makes us overconfident

2:23 thinking we know more than we do I've seen many solutions that fall short are overly complex and use unnecessary resources if you took my machine learning course you know what I'm talking about there we use the huge neural network to get the best results but one with the fraction of the size could have good accuracy as well and we could explain what that does so I think there's real value in revisiting Basics from time to time throughout the course I'll give you homework assignments to improve your logic and deepen your understanding I'll also give you a final challenge toach teach the car how to race instead of following the rules nicely I'll host a live stream event where I race against your AI cars and there will be prizes I just haven't figured out what those are yet but stay tuned and while you're designing your race car I'll continue to teach how to code some things as well for that part you do need to know some math and JavaScript this playlist can help with that and we will continue the self-driving car project so it's good to be somewhat familiar with it but will mostly just add new functionality into it so if you feel confident in your

3:39 skills take the last version from GitHub and try to follow along what I'll teach then is how to implement drra shortest path algorithm so the car knows how to reach its destination I'll also teach how to make the game mechanics we'll control the main car but the others will be AI I'll show you how to monitor their progress and make the scoreboard now this top view is not great when racing I think I'm really used to seeing things from the car's perspective so I'll teach you how to code the camera sensor where we render what the car sees this might be useful someday for object recognition but for now I just like it and what I like even more is this view from behind the car I'm going to teach you how to do that as well and I'm not done controlling the car with the keyboard is not ideal I'll teach you how to implement analog steering Twice first by turning this into a mobile app and using the device orientation sensor to turn I think it's better than using the keyboard but I have an even better one using the camera some basic image

4:49 processing and these blue wristbands we're basically going to become Iron [Music] Man oh yeah and the sound it's procedurally generated from scratch all of this is just plain JavaScript no libraries like everything else on the channel so you can learn all inner workings of a complex system like this and all of this is AI the new camera sensor the path finding the fancy controls using the smartphone or this thing image processing augmented reality they all make the system more intelligent that's what I want you to get from this course AI is often a combination of things not just neural networks and artificial intelligence is not the same as machine learning many people confuse the too there's no machine learning in this course which reminds me someday I'll teach you how to add machine learning into the system as well but for that we need data so if you want to help go to this link and raise try to beat my time from there if you make an account and do

5:58 a good job your name will appear there and others will see you racing next to them that's actually me racing I'm recording every move we do so I can replay it like that but don't worry about making mistakes for machine learning we need to teach the car what not to do as well so mistakes are more than welcome excited great now get ready to put your neurons into overdrive this is the playground and look at the car it's doing something it's applying the so-called right hand rule for solving mazes so if you're inside a maze put your right hand on the wall and you'll eventually get out of it most of them anyway there are some mazes where it doesn't work but we'll get to that later now just look at this the car is exploring it's going everywhere and it's actually great it's not easy to do this you can press this manual override button on the right and now the car is controlled by the keys on the keyboard and just try to use the arrow keys and