>>Okay, we are here in the Google Self-Driving Car. We’re inside. It’s obviously driving itself. And it’s doing everything by itself.>>And I am here as a passenger. I’m really hoping that Sebastian did a good job in training this, because he is not driving right now, and I->>I am not driving, it drives better than me.>>I am at the mercy of this car. That’s true, it does drive better than me.>>So why are we talking about self-driving cars then?>>Why are we talking about self driving cars? Well we’re going to start by talking about supervised classification, and self-driving cars are one big supervised classification problem.>>What does supervised mean?>>Supervised means that you have a bunch of examples, where you know sort of the correct answer in those examples. So I know that you have an example of this from the self-driving cars.>>Yeah so I mean we train our car and we show the car what’s the right behavior. And we did the DARPA Grand Challenge we would take it out for a spin and it would very carefully watch us human drivers drive and would emulate our behavior.>>And in fact, this is sort of how people learn to drive, right?>>Yeah, I think so. When I was a child, I watched my parents drive. And they weren’t really good drivers, but I would still copy a lot of stuff from them.>>Yeah, so it’s kind of like, in the way the humans drive by watching lots of examples. That’s what computers do when they’re doing machine learnings. You give them lots of examples, and they start to figure out what’s going on. That’s what we’ll be learning about.>>And it’s totally true. In this unit, you’re going to be learning about machine learning, the same way we program self-driving car. You’re going to program data and test out whether you can make a car go fast and slow at the appropriate time. using machine learning supervised learning.>>That’s right. So we’re going to be looking at a really cool terrain classification problem that was very important for Stanley. You want to introduce that?>>So, in Stanley’s case, I was driving through the desert and the desert terrain is like ruts and broom and can be very brutal. So it you drive too fast, you are running the risk of flipping over and destroying yourself. So one thing we’ve trained the car to do is to really slow down at the appropriate time. We did this not by writing little rules, we did it by us demonstrating to the car how we drive. And it will just emulate us.>>How many miles did you have to drive to train that.>>Oh thousands of miles. We spent thousands of miles everyday in the desert.>>Wow.>>And it took quite a while for it to become smart.>>Your poor grad students. I can only imagine.>>Well, I was the guy who had the pizza for everybody, but, it was a great time because it was no email, we just had the, us and the software. And every time you got the software back it was very obvious, the car would punish us.>>Oh, that sounds great. So, I think we should probably get started with that. Let’s try out a few different supervised classification problems.>>Yeah, so the unit’s all about supervised learning, so let’s dig in.>>Sounds great.