1.1 Machine learning lecture 2  (Page 2/15)

So I just want to start by showing you a fun video. Remember at the last lecture, the initial lecture, I talked about supervised learning. And supervised learning was this machine-learning problem where I said we're going to tell the algorithm what the close right answer is for a number of examples, and then we want the algorithm to replicate more of the same.

So the example I had at the first lecture was the problem of predicting housing prices, where you may have a training set, and we tell the algorithm what the "right" housing price was for every house in the training set. And then you want the algorithm to learn the relationship between sizes of houses and the prices, and essentially produce more of the "right" answer.

So let me show you a video now. Load the big screen, please. So I'll show you a video now that was from Dean Pomerleau at some work he did at Carnegie Mellon on applied supervised learning to get a car to drive itself. This is work on a vehicle known as Alvin. It was done sort of about 15 years ago, and I think it was a very elegant example of the sorts of things you can get supervised or any algorithms to do.

On the video, you hear Dean Pomerleau's voice mention and algorithm called Neural Network. I'll say a little bit about that later, but the essential learning algorithm for this is something called gradient descent, which I will talk about later in today's lecture. Let's watch the video. [Video plays]

Instructor (Andrew Ng) :So two comments, right. One is this is supervised learning because it's learning from a human driver, in which a human driver shows that we're on this segment of the road, I will steer at this angle. This segment of the road, I'll steer at this angle. And so the human provides the number of "correct" steering directions to the car, and then it's the job of the car to try to learn to produce more of these "correct" steering directions that keeps the car on the road.

On the monitor display up here, I just want to tell you a little bit about what this display means. So on the upper left where the mouse pointer is moving, this horizontal line actually shows the human steering direction, and this white bar, or this white area right here shows the steering direction chosen by the human driver, by moving the steering wheel.

The human is steering a little bit to the left here indicated by the position of this white region. This second line here where Mamos is pointing, the second line here is the output of the learning algorithm, and where the learning algorithm currently thinks is the right steering direction. And right now what you're seeing is the learning algorithm just at the very beginning of training, and so there's just no idea of where to steer. And so its output, this little white smear over the entire range of steering directions.

And as the algorithm collects more examples and learns of a time, you see it start to more confidently choose a steering direction. So let's keep watching the video. [Video plays]