Andrew Ng is a big name in the field of 机器学习 and a professor at Stanford University. I actually got to see him in the flesh!!

This afternoon, Biao from the lab dragged me out of the lab, saying a big shot was coming to give a talk at 3:30 p.m. Go grab seats, quick.

I checked the time — 2:30. What's the rush?

Having been holed up in the lab all day, I had no idea which "bull" was coming or what the talk was about. Biao said it was 大数据, machine learning and stuff. Right up my alley. Whoever it was, a Jersey cow or a Holstein, it's all good beef.

Then the moment I walked into the classroom, it turned out every seat was already taken, with onlookers standing everywhere. I squeezed to the front and squatted in a corner. Then came an hour of waiting.

15:31, the talk officially began, and that's when I realized it was this particular bull. I was instantly delighted — sitting on the floor for an hour and nearly getting a cold butt hadn't been in vain.

He adorably opened with two sentences of Chinese, then switched to English and started flipping through slides.

First, he compared using two certain positions on a motorcycle as features versus choosing the front wheel and the handlebars as features, to illustrate the importance of picking the right features.

Then, as the title said, "via Brain simulations," he started from human perception and explained the theory of machine learning.

Then came two points about machine learning:

It's not who has the best algorithm that wins. It's who has the most data.

Large numbers of features is critical. The specific learning algorithm is important, but ones that can scale to many features also have a big advantage.

Honestly, all pretty standard stuff — a pity that clever folks often forget these things.

Then he moved on to applications of machine learning and so on.

Most of the content was about the same as this PDF of the slides I found online, with some revisions. (Please don't download it with Xunlei, it hurts...)

For example, the last slide about Deep Learning had been slightly revised. This time the content was:

  • Deep Learning Lets learn our features
  • Discover the fundamental computational principles that underlie perception.
  • Scalling up has been key to achive good performance.
  • Recursive representations for language.
  • Online tutorial (China)...http://deeplearning.stanford.edu/wiki/

PS: I took the chance to look up the papers I've been reading. It turns out my earlier belief that "a good chunk of my references were written by him" was wrong — there's another guy named Andrew McCallum in the CRF field. Between the two of them, they wrote half of my machine learning reference materials.

Also, to be honest, I really just went to see his face — kind of like reading Fortress Besieged and then getting to recognize Qian Zhongshu's face. The talk was fun, but in the end it's still papers and courses that really count.