#aiclass versus #mlclass - fight!
I've been doing Stanford's online Introduction to Artificial Intelligence and Machine Learning courses. Both have their good points and bad points: here are my comments so far.
[The Introduction to Databases course looks interesting too, but I only have a finite number of hours in the day!]
Calibration: I have a first-class undergraduate degree from Oxford and a PhD from Glasgow, both in pure mathematics (mostly algebra/category theory). I've taught first- and second-year undergraduate mathematics. I've spent about four years working as a professional programmer, but I've been programming for long before that as an amateur. I have some limited dabblings in ML, and none in AI.
Logo: Friendly robot in a mortar-board versus scary cyborg with glowing red eyes. MLClass wins this one hands-down. This may sound like a trivial point, but I find it makes a surprising difference to how keen I am to visit each site.
Expected background: I've been really surprised by how unwilling both courses (but especially MLClass) are to assume understanding of what I consider to be high-school mathematics - matrix multiplication, log functions, conditional probability, simple calculus. Aren't Stanford undergrads expected to know this stuff? On the other hand, not assuming much mathematical background makes the material accessible to more people, furthering their (very noble!) goal of opening up first-class education, so I'm willing to give them a pass on this one.
Pace: Both courses are a lot slower than I'm used to from Oxford and Glasgow. AIClass is going faster, though, so this round goes to them.
Technology: MLClass easily wins this one. Their videos are available for download and off-line viewing, can be viewed at different speeds, and don't start until you're ready for them. They have a nifty system for submitting programming assignments. AIClass' reliance on YouTube means that they're inaccessible in countries which block that site. The AIClass servers have gone down every time a homework assignment has been due (though to be fair, they have over twice as many students as MLClass).
AIClass: please adopt the MLClass technology if you run the course again next year.
Quizzes: I really like the way both courses use quizzes embedded in the videos to test your understanding of the material. The AIClass quizzes are more frequent and more testing, but I'm going to give this round to MLClass because they make sure each quiz shows enough information on-screen to answer the question. With AIClass I usually have to re-wind the video several times in order to make sure I fully understand the problem.
Programming exercises: MLClass has programming exercises, AIClass doesn't (though you're encouraged to write your own programs to help you answer the homework assignments and quizzes). Point to MLClass.
Lecturers: I really like Andrew Ng's lecturing style in MLClass. He's bright and engaging. For AIClass, Sebastian Thrun is pretty good, but Peter Norvig (while obviously a giant in his field) reminds me inescapably of Ferris Bueller's economics teacher. Point to MLClass.
I make that 5-2 to MLClass. But both courses are fascinating, and it's exciting to be involved in such a bold experiment in distributed learning.
[The Introduction to Databases course looks interesting too, but I only have a finite number of hours in the day!]
Calibration: I have a first-class undergraduate degree from Oxford and a PhD from Glasgow, both in pure mathematics (mostly algebra/category theory). I've taught first- and second-year undergraduate mathematics. I've spent about four years working as a professional programmer, but I've been programming for long before that as an amateur. I have some limited dabblings in ML, and none in AI.
Logo: Friendly robot in a mortar-board versus scary cyborg with glowing red eyes. MLClass wins this one hands-down. This may sound like a trivial point, but I find it makes a surprising difference to how keen I am to visit each site.
Expected background: I've been really surprised by how unwilling both courses (but especially MLClass) are to assume understanding of what I consider to be high-school mathematics - matrix multiplication, log functions, conditional probability, simple calculus. Aren't Stanford undergrads expected to know this stuff? On the other hand, not assuming much mathematical background makes the material accessible to more people, furthering their (very noble!) goal of opening up first-class education, so I'm willing to give them a pass on this one.
Pace: Both courses are a lot slower than I'm used to from Oxford and Glasgow. AIClass is going faster, though, so this round goes to them.
Technology: MLClass easily wins this one. Their videos are available for download and off-line viewing, can be viewed at different speeds, and don't start until you're ready for them. They have a nifty system for submitting programming assignments. AIClass' reliance on YouTube means that they're inaccessible in countries which block that site. The AIClass servers have gone down every time a homework assignment has been due (though to be fair, they have over twice as many students as MLClass).
AIClass: please adopt the MLClass technology if you run the course again next year.
Quizzes: I really like the way both courses use quizzes embedded in the videos to test your understanding of the material. The AIClass quizzes are more frequent and more testing, but I'm going to give this round to MLClass because they make sure each quiz shows enough information on-screen to answer the question. With AIClass I usually have to re-wind the video several times in order to make sure I fully understand the problem.
Programming exercises: MLClass has programming exercises, AIClass doesn't (though you're encouraged to write your own programs to help you answer the homework assignments and quizzes). Point to MLClass.
Lecturers: I really like Andrew Ng's lecturing style in MLClass. He's bright and engaging. For AIClass, Sebastian Thrun is pretty good, but Peter Norvig (while obviously a giant in his field) reminds me inescapably of Ferris Bueller's economics teacher. Point to MLClass.
I make that 5-2 to MLClass. But both courses are fascinating, and it's exciting to be involved in such a bold experiment in distributed learning.