Hello, this is John Moore, professor at Princeton University. Nice to see you today. Well, in fact I can't see you, but still welcome. We're going to discuss a machine learning today, in particular an introduction to the area of machine learning. Now, this topic has gotten quite exciting over the last few years. What has caused the excitement? Part of it is the fact that many jobs that we've seen have been transformed by computers. If you go to an auto assembly line, you see robots that are taking over and doing much of the job of what an individuals had done in the past. Similar thing happens in computer facilities. Most of the computers are made by computers nowadays. This area has been transformational. Second, it has to do with digital assistance. If we were driving over to your work on a particular day or going to the beach on another day, you'd find that you go to Google most times to see if the traffic goes bad in a particular area, and you'd see that perhaps, there's a better route. Ten years ago this would have been quite revolutionary, who just would not have expected that? This type of digital assistant goes back into the realm of science fiction. A recall is a 10-year I'll go into a movie where there was a robot, Robbie the Robot, and that robot was in a movie that the robot was your friend and could help you. We've seen that in many other films, in Star Wars, Black Panther more recently, we see these digital assistance taking over as almost friends for individuals. Can that happen going forward? What will that mean for our society? How many jobs will be changed over time? So this is part of the excitement that we see in the area of machine learning. Also we've seen the notion that computers will be driving cars for us. Who could have imagined that a few years ago? It just seems almost impossible. So let's take a look now at the notion of a large population, and let's see if we didn't identify what are the characteristics in that population that might be of interest to a particular problem? For example, who in that population? We see a large population here, who in that population will be given the diagnosis of a cancer or not? How can we identify those people in the population that have certain characteristics? This is a goal of machine-learning. Now, there are two types of machine learning that we're going to look at in these modules for this online course. First one is called supervised learning. In this area of supervised learning, we have labels that, in fact, answers to those questions. So we have the population which you think of as a large adequate set of data, and we've identified those people who have certain characteristics. The machine learns to identify what are the features that lend themselves to identify those people who have this characteristic. The second thing that we're going to look at is called unsupervised learning. In this case, we're going to look at no longer labels, we're not going to require labels, we can have the computer tell us what are the patterns that arise in that context. Now, what are the origins of machine learning? If you go back into the 1950s, and you went to Princeton in the statistics department, you would've met John Tukey, who was the founder of a topic called Exploratory Data Analysis. This topic was the origins of data science. In fact, I spent many years with John, several years rather, working with him on some projects. He would look at data for hours and hours on hand and bring data from the Treasury Department to his office, and he would pour over the data for hours, looking to see what the story was behind the data? What is the data telling us? That's really the origins here of machine learning. Can we learn something in the data which tells us something interesting? So that's part of the origins. Of course, we now have much more data today than we did back in the '60s and '70s. What's sometimes called massive data. When I'm going to do is call that adequate data, we're not going to necessarily say it's massive, we're just going to say it's adequate or not adequate for the task at hand. Here we see another application of machine learning. It's called facial recognition. This is my granddaughter who is one our old in this first picture, then we see her at five months, and we see her now at 13 months, which is the current age in San Francisco. The computer can identify these characteristics not just at one hour old, but all through her lifetime. If she's lucky, she'll be living to a 100 years old. We want to have an adaptive system that can identify those features which allow us to identify this person over time. We're going to see similar ideas in the context of economics. Another application which seems to be a revolutionary is the topic of computers playing games. A number of years ago we found that computers can play chess better than any human has ever played chess. Last year in 2017, we saw that AlphaGo, the Google system, beat the best Go player in the world. This was quite revolutionary. People thought for many years that it would take decades before computers could be the best Go player in the world. Last year did it, and it was quite a revolutionary in this area. So the take home points for today are as follows. First of all, we've seen much excitement in the area of machine learning. What is it that allows us to teach a computer how to learn? We want to identify that. What are the characteristics that we've seen is that individuals learn differently than computers. In fact, my granddaughter Eve as she learns, she's a learning machine, humans are learning machines, not machine-learning, different characteristics. Secondly, we now see much more data available, and we're going to identify the idea of availability of data would be an adequate. You could call it that massive data, but we're going to call that adequate data. Finally we're going to understand the difference between supervised learning where we have labels, it tells us right or wrong and those are one areas of much success. We're going to identify that compared against unsupervised learning where we see no labels but we're looking for interesting patterns in the data. What is it the data is telling us? What's the story underneath that?
