Here are my thoughts, in no particular order:
- It was a very good course. I learned a lot of useful information in a format that was much easier to digest than if I'd just read a textbook.
- There were ten weeks and each week had a series of lecture videos ranging in length from 5-10 minutes. There were in-lecture quizzes (which didn't count for anything) as well as comprehension quizzes to be taken online after the quizzes. Finally there were homework assignments that required writing code in Matlab or Octave. Solutions were uploaded and graded automatically through a nice seamless interface.
- I loved that the lectures could be watched online at variable speeds. I also loved that I could download the videos and watch them on my phone during my subway commute to and from work. Being able to pause, skip, and review video is a huge advantage. If you zone out for a moment and miss some important point, you can just re-watch it instead of being confused for the remainder of the lecture. Win. I also loved that I was able to download all the slides.
- The quizzes were often pretty challenging. Mostly they were multiple choice but sometimes I had to do numerical computations.
- The homework assignments were often interesting but in order to make them tractable for large numbers of students of many different backgrounds and strengths, the instructors provided an awful lot of pre-written code. This made many of the assignments feel a little plug-and-chug, even though they were tackling pretty complex real-world issues (e.g. spam detection).
I think that last issue is the real sticking point for MOOCs. For a class like this one to really be excellent, students must be challenged with big open-ended design problems that can be solved with many approaches. The student has to figure out for example, what are the best feature vectors, how to code them, how to tune the parameters, etc. Those design problems must be tough and force the student to really struggle. That in turn requires instructors or teaching assistants that can work hand in hand with the students, answering questions, making suggestions, and so on. While the course had an online message board that was supposed to play this role to some extent, I don't think its the same has having face time with an experienced instructor.
None of this is said to knock Coursera or Dr. Ng. Quite the contrary, I feel like I got a great bargain on my investment of zero dollars (although I put a good deal of time into it). However I'm just thinking about the bigger picture of how a MOOC would have to look to be truly a replacement for brick and mortar universities.
I think another big problem MOOCs will face is that a whole lot of students need advising - what courses to take in what sequence, how to study, how to do problem sets, etc. While many individuals can sort this for themselves, a lot of folks can't, and thats a big part of the value that a university provides. Right now I think MOOCs are probably a great resource for people who already know how to study or already have some higher ed under their belts. This article from Inside Higher Ed has some interesting demographic information on who takes MOOCs.
Anyways, it was a fun course and as far as the instructor is concerned, I am now an "expert" (his words!) on Machine Learning!