Embracing Uncertainty
Christopher Bishop | 32:34
Transcript
I work in the field of machine learning, I'm going to tell you about some of the exciting developments that are happening in this field. And really, it's all about dealing with uncertainty, which is an interesting problem for computers, because computers, fundamentally, as you all know, are very certain things in the sense, they're based on logic. Everything is zero or one, everything is true or false. It's completely certain. And yet the world in which computers operate, the data that computers have to deal with is full of uncertainty. So just have a little look at some examples of what I mean by uncertainty in the context of computing. So here's a picture of an object, in this case, an animal alarma.
And one of the things that we want to do with images sometimes is to cut out the objects or remove the background. And in this case, we might try to solve the problem by looking at colors. So the difference between a foreground object and a background often lies in in color and texture. And the problem with with this particular example, of course, is the animals quite well camouflage. There's a great deal of uncertainty about which is foreground and which is background. And yet, we need to address that uncertainty. If we want to remove the background. We'll come back to that in a moment. It's another example that we'll also return to two people are going to play chess, which person is going to win the game? Can we predict the winner?
Again, it's full of uncertainty, we might know that one player is stronger than another player, but it doesn't mean that the stronger player will necessarily win. So again, there's uncertainty. There's a very practical example. This is a web page, it's a page of search results from Microsoft search engine, Bing, and somebody has typed in LCD into the search box. And here are some search results. The first one is a Wikipedia page and so on. But also we have these other links. And these are advertisement. So that's how search is paid for. So the model is that advertisers bid in an auction to have the advertisement shown alongside the search results, and they bid against these keywords. So people are interested in selling LCD televisions, for instance, will bid to have that advert shown against a search like LCD. And if they win that auction, then their adverts get shown.