1.14 Machine learning lecture 15

MachineLearning-Lecture15

Instructor (Andrew Ng) :All right. Good morning. Welcome back. So a couple quick announcements before I go into today's technical material. So unfortunately, a long time ago, I agreed to give a few [inaudible] in Japan, so I won't be here on Wednesday. So the next lecture will be, I shall be taping it ahead of time. In particular, instead of a lecture on Wednesday, I'll be giving Wednesday's lecture tomorrow instead at 1:30 p.m. in the Skeeling 191 auditorium.

So those here, you come to class at the usual time, the taped lecture will be taped ahead of time, but then shown in this room at this time. But if any of you are free tomorrow at 1:30 p.m., I should be grateful if you can come to that, alternative of the taping of the lecture, which will be at 1:30 p.m. Skeeling 191.

I'll also send an email out with the details, so you don't need to write it down. But if any of you can make it to that, it'll make it much easier and more fun for me to teach to a live audience rather than to teach on tape to an empty room. So I'll send an email about that. We'll do this for this Wednesday's lecture as well as for the lecture on the Wednesday after the Thanksgiving break.

In case you don't want to deal with it, and this is the only time you make it, you can just show up at the usual time, and the lecture will be presented here in this room as well, and online, of course. Let's see. For the same reason, my next two office hours, which are usually on Fridays will also be moved to the following Monday because I'll be traveling. So if you're planning to come to my office hours either this week or the week after Thanksgiving, just take a look at the course website for the revised time.

In case you want to meet with me, but can't make the new time on the course website, you can also send me an email. We'll try to set up a time outside that. Okay?

But again, if you're free at 1:30 p.m. tomorrow, either those watching this online now or people in this classroom, I would be grateful if you can come to Skeeling 191. If there are any of you that couldn't ever make this regular class time and so always watch it online, you can come see a live lecture, for once, those of you watching online.

Are there any administrative questions about that? Okay? Let's go into today's technical material.

Welcome back. What I want to do today is continue a discussion of principal components analysis, or PCA. In particular, there's one more application that I didn't get to in the last lecture on [inaudible] indexing, LSI. Then I want to spend just a little time talking about how to implement PCA, especially for very large problems. In particular, I'll spend just a little bit of time talking about singular value decomposition, or the SVD implementation of principal component analysis.

So the second half of today's lecture, I want to talk about the different algorithm called independent component analysis, which is, in some ways, related to PCA, but in many other ways, it also manages to accomplish very different things than PCA. So with this lecture, this will actually wrap up our discussion on unsupervised learning. The next lecture, we'll start to talk about reinforcement learning algorithms.