|
I've watched the first few PGM lectures from Open Classroom. I noticed that the way things are presented there is less compelling than in the AI course. This has nothing to do with Daphne Koller's teaching as I'm enjoying that so far. Rather it has to do with the format of the course and the way that text appears. There are two primary images in the PGM lectures (1-4). First, a headshot of the instructor in the lower left-hand corner (web-cam quality) with the other three quarters of the screen used for text. Second the "magical appearance" of text on the full screen. The second one matches the way that text appears in Khan Academy lessons except that it's black-on-white instead of many-colors-on-black. In the AI lectures the two primary images were full-screen headshots - Sebastian or Peter talking directly to the camera (and so to me/you) and writing on paper with a human hand and a pen viewed from behind the writer. I think the AI version is better and more personal. The PGM model has me looking at a computer monitor with an invisible someone writing text from behind the screen. The AI model has a human writing in ink on a piece of paper in front of me. This is much the same feeling you get when trying to explain something to friend in a bar or restaurant. You grab a (paper) napkin and start drawing/writing. I thought of the AI teachers as Peter and Sebastian almost immediately. It seems unlikely that I'll think of the PGM teacher as Daphne unless the lecture model is different. Your thoughts? @robrambusch ♦ Could you please post the link for the open classroom PGM class? I only know of http://openclassroom.stanford.edu/MainFolder/HomePage.php but PGM isn't listed there.
(03 Feb '12, 13:40)
octon
Not to worry... found it. http://openclassroom.stanford.edu/MainFolder/CoursePage.php?course=ProbabilisticGraphicalModels
(03 Feb '12, 13:41)
octon
|
|
I'm more to the side of @robrambusch "feelings", I'll try to explain my point congruently, hopping to be succesfull. |
|
For me the flow of "concepts" in ML and now also (after doing 5 videos) in PGM were much more cohesive and absorbable over the 8-12 minute uninterrupted videos than the 1 to 2 minute ones of AI, interrupted with quizzes. The separated review questions in ML also meant you could absorb the whole topic and then do the questions, submit them and get immediate feedback on where you went wrong (say 3.75/5.00). You could then go back to the topics, make sure you understood and try the review questions again. But here's the rider - the questions in each attempt would be different, so you might even get less marks in the second attempt! This means that you are really motivated to get 5.00/5.00 and by the time you do after perhaps 3 attempts, you REALLY understand the topic. I found by the end of the course I was making every effort to absorb the videos so I could get 5.00/5.00 first time round. The homework programming assignments were similar with online submission and immediate feedback and resubmission. By the time you got 100% for the homework, you pretty much understood how to implement the algorithms. Here's hoping the actual Coursera PGM course adopts the same review question and homework assignment processes! |
|
Speeding up the ML videos made them less engaging but more convenient. We had a choice of 1.0x, 1.2x, or 1.5x. I ended up preferring ML's format over AI's, but that was more a matter of attitude than the interface. So I'd be happy with either format, as long as hands don't cover up the writing too often. |