What class would you teach?

Sep 10, 2008 13:53

So when potential hires apply for faculty jobs, one of the questions they're asked is "If you could teach any class you wanted, what would it be ( Read more... )

Leave a comment

Comments 12

skydiamonde September 10 2008, 21:16:08 UTC
My year there was 16-299 - Intro to Feedback Control Systems that was geared towards computer science majors.

http://www.tcpulse.com/psched/course/16299/

It was taught by Yoky (http://www.cs.washington.edu/homes/yoky/)

I sucked and barely got a D. It was seriously be like a point or two on the final exam. Before going into the final, I went crying to Mark Stehlik trying to decide if I should just pull out and get the W on my transcript and retake it semester, but he convinced me that I might as well try the exam. They did teach PID control, out of the list of things you said you wouldn't cover. I'm not sure I remember what else was in the course.

Reply

r_transpose_p September 10 2008, 21:29:12 UTC
Cool. It makes sense that CMU would teach a course like this.

It looks like my "semester 1" is the same thing but without the "classical linear control" Perhaps the course would've been easier without this?

(Note : I took my controls classes in exactly backwards order, starting with non-linear control, than taking a multi-input-multi-output class on linear control where everything was presented in state space, then finally taking a classical controls class. I still found the classical controls class to be the least intuitive and the most frustrating (although non-linear was the most difficult -- imagine 15-251 level difficulty, and pencil-and-paper assignments that take as long as the projects for most non-systems programming classes)

My second choice would be to teach a class based on:
http://tintoretto.ucsd.edu/jorge/dcrn/index.html
which either my proposed 1st-semester course, or 16-299 would be a decent prereq for.

Reply

r_transpose_p September 10 2008, 21:43:35 UTC
The funny thing is as I was writing the syllabus I was thinking "Huh, this seems a bit much for one semester. CMU undergrads could probably handle it, but it'd be rough"

And my course wouldn't even cover some of the stuff in Yoky's course.

(the stuff I'm omitting from semester I is pretty much "The stuff you don't need for my research" The stuff in semester II is "stuff that is more useful for robotics than anything from classical controls" You would not be able to get a job as a controls engineer after taking the sequence I propose. You probably wouldn't even be able to understand what most controls engineers talk about. But you would be able to do cutting edge research in robotics or controls. Interesting how that works sometimes. An analogy to what I'm trying to do might be "catching mathematicians up to speed so they can do PL research" They probably wouldn't be able to write an operating system afterwards.)

Reply

skydiamonde September 10 2008, 21:50:37 UTC
Heh I bet I'd go better in your course PLUS it'd probably make more sense to me.

They also did this weird thing in the first few weeks that were mini-crash courses on random physics and diff eq stuff that CS majors aren't required to know (I actually managed to never take diff eq). It was weird cuz the way it was done, it had nothing to do with anything at first and they were completely disjoint topics that would come together later, but seemed completely random in those weeks.

Reply


soong September 11 2008, 04:46:15 UTC
At CMU, to get my minor in robotics I took MechE Controls because ECE failed to teach it 3 years in a row and everyone had to take MechE version because it was the only game around. In the end it was mostly math and the applications the put in the word problems didn't matter ( ... )

Reply

r_transpose_p September 11 2008, 06:24:49 UTC
Yeah, I should try building some actual robots before saying "This is the way to teach it to computer scientists, you don't need the other stuff ( ... )

Reply

r_transpose_p September 11 2008, 06:28:43 UTC
Cool fact : Those "dynamic programming" algorithms we learned in classes like 15-211 were originally invented to do optimal control.

http://en.wikipedia.org/wiki/Bellman_equation
for the discrete case

and the article on the continuous case is harder to read, but its really the same thing -- especially since everyone numerical solves the resulting equations anyway!
http://en.wikipedia.org/wiki/Hamilton-Jacobi-Bellman_equation

Reply


szkoda September 12 2008, 04:23:05 UTC
That sounds like a pretty awesome class. If you taught it at a school near me, I'd love to audit it.

Reply

r_transpose_p September 12 2008, 21:25:57 UTC
Hehe, if you've taken "dynamics" from a math department (or an equivalent class) you would ace the first part.

In fact, you could probably get the same material by looking at whatever college you're at, checking whether the mechanical engineering department has a grad class on non-linear control, and auditing that.

The parts that build on "the rest of control" are pretty self-contained, and the rest of the class doesn't really hinge on them. Plus, if you have a differential geometry background, most of the material (for a real course on the subject) will be much easier for you than for the engineers.

Reply

r_transpose_p September 12 2008, 21:32:15 UTC
In the last paragraph I'm talking about "graduate non-linear control as it is usually taught"

After taking that, you can look to see if anyone lets me teach "course II" in the sequence. Probably the closest thing out there is Andrew Ng's machine learning class at Stanford:
http://www.stanford.edu/class/cs229/schedule.html

Reply


Leave a comment

Up