Re: Wow, what a ride!
A little belatedly I want to join the large group of people thanking Caltech for making this class publically available and Yaser for his fantastic job in teaching it.
I'm in industry, hoping to apply what I have learned on a couple of upcoming projects this year. I found the course content extremely worthwhile.
I find Machine Learning difficult, because it is counterintuitive to me. Even after the class, I find it hard to rid myself of the notion of applying all possible cleverness to somehow explain the available data and then to extrapolate that explanation, i.e., as we learned, the notion to minimize the insample error without regard to the outofsample error. And the Netflix example, to predict people's preferences for movies without any input of domain expertise on movies or psychology, that's still black magic, even after understanding how it's done.
While it's easy to pick up a couple of algorithms from the many textbooks and online materials out there, it is the solid foundation, both mathematical and practical, as well as this better intuition that I would have missed studying alone without this class. Also, the cadence of the lectures and the homeworks helped to enforce some learning discipline.
Regarding the details of the course content and their overlap with my planned projects, I would have liked some more discussion of other practical problems. One is methods to handle classification problems where one result occurs much more frequently than the other. Another is methods for data sets that are not well described by tabular x and y's, but more naturally through a graph structure.
On the other hand, I cannot really think of any topic I would have liked to have been dropped from the course. And I imagine that the unhurried pace of the lectures was one of the magic ingredients that made the course so successful.
In summary, not many suggestions how to improve a class of such high caliber. Thanks again.
