Yaser Abu-Mostafa was (by enormous margin) the most effective professor I had at Caltech. Despite being such an expert in the field, he understands clearly when a concept is particularly challenging--and what about it makes it so. This class (the official equivalent) was one of my absolute favorites. Definitely worth a look!
What Andy says is absolutely true. This machine learning class was easily the best class I took at Caltech. Prof. Abu-Mostafa got a standing ovation at the end of the course the term I took it. I wish I could have taken more of his classes.
It was also fairly difficult -- the assignments were hard, but at every step, you could look at what you'd done and say "I know why I'm doing this, and I can see how this works."
I remember at the end of the term he took several students' notes and made copies of them, so that he could compare the students' notes with what he was trying to convey, and could know if he wasn't teaching certain parts of the class well enough.
It's a shame that not all professors are as dedicated and responsive to their teaching obligations as Prof. Abu-Mostafa.
Oh, also, "introductory" in this context is meant to differentiate it from "graduate level". Every student (mostly juniors and seniors) in this class will have had several terms of math, theoretical CS, and practical programming classes.
This looks like the kind of class more people need to see. It's less "how to implement" things and more why things do or don't work.
I find this dreadfully important because when you study this math you realize that more often than anyone expects, standard ML is extraordinarily fragile, but also has some powerful justification. For instance, this[1] made me laugh with joy.
I agree that understanding the theory is very important.
The figure is very interesting. Would you care to explain it? I think I know what these are in theory, but perhaps I haven't internalized them enough to understand the visual representation.
Online teaching is not so much about watering down than understand that your audience will not usually and probably cannot have the same level of focus that your on-campus students will have. It's not a judgement of skills, but a simple observation of psychological incentives.
Online is a great place to learn, but it's absolutely the wrong place to learn the exact same curriculum as offline.
These seem to be LIVE videos of the lectures broadcast during the workday in US.
Their previously recorded page states that they will only provide videos of the first week.
Not really doable by anyone in US timezone with a job.
the best aspect of the andrew ng course was the homeworks using octave. Yes, it's watered down and not as mathematically rigorous as the real course, but you learn a lot of the essentials from the experience of coding machine learning algorithms, that I can't imagine learning as easily from doing non-programming homework.
Prof. Ng's class wasn't watered down. CS229A(http://cs229a.stanford.edu/) is the Stanford equivalent of the online ml-class. Ng also teaches another machine learning course at stanford(CS229) which focuses on the theoretical underpinnings of ML.
Having taken the online course, I'm not sure I understand you - the programming assignments in some cases required 1 line of code be written (in the middle of a more complex program). I find it really hard to believe that students who pay tuition would be given the same assignments.
CS229A is a course taken by people from different backgrounds not just CS. It basically deals with the practical aspects of machine learning, implementation issues etc. In addition to the lectures and the assignments, stanford students also had an additional course project.
My favorite classes (and the ones I've learned the most from) have always managed to balance theory and implementation. The formula has usually been a mix of
1) purely theoretical exercises emphasizing fundamental concepts
2) project-based assignments in which you must understand the theory and write a decent amount of code to apply it. Usually a bunch of code not central to the concepts has already been written for you. But just implementing it is still not enough. To test your conceptual understanding, they ask you to run your code in various situations and explain the results.
I dunno. You're given the equations in that class. Translating from math notation to octave syntax doesn't require you to have any idea of why any of it works.
I think it's actually quite telling if other institutions feel obliged to ridicule efforts (by Coursera et al) to make online learning a new experience, rather than just copying existing concepts as exercised in traditional universities.
As an undergrad at Caltech who will be taking this course next term, I find this intriguing. I'm wondering if this course will be as popular as the Stanford courses (looks like a lot less effort is being put into organization, design, etc) and how the difficulty of this will compare to the Stanford courses and an average Caltech course.
Homeworks? Lack of a sign-up link? Interesting that they have grammatical errors on this page considering their goal here (to compete with other top universities doing the same thing)
who says they are trying to compete with other universities? the goal here is to provide quality education accessible to everyone. it also looks like the webpage was created by the professor himself (http://work.caltech.edu/ is his personal page) and may not be completely ready yet.
There is a whole subsection of the internet dedicated to exactly that. Mini-sites are generated either using wordpress or other scripts get content from all over the internet, spin it and then put it up on a domain with a standard template and stick adsense advertisements on the side. Makes pretty good money if you have a large enough network and do some SEO.
The course seems all over the place. You learn very little about a lot of things... The topics don't really seem to build on each other. I'm not sure that's a good thing.
In total it might be a lot, but it can't be indepth.
for example "Error and Noise" is something you can spend a whole year studying. What are you really going to cover in one lecture? You'll just touch on a few things that might be useful at some point, but you're not building a larger encompassing understanding of "Error and Noise".
I've taking classes like this. You learn a few useful tricks, but when shit hits the fan in the real world and your tools are not enough you're left floundering. You can't prove things, you can't develop your own methods because you don't understand the principles they're based on.
This would probably be a cool class to take freshman year so that you can figure out for yourself what you'd like to study.
This is clearly an introduction. You're not going to spend a semester talking about "Error and Noise" in a class like this. I had the opposite impression --- that it covers a relatively small number of topics that are very highly related.
Now, will you come out of this class knowing how to do practical machine learning? Probably not. But these are fundamental concepts that you must know. When your tools are not enough in the real world, you can appeal to these concepts to understand why.
Is there any reason to watch this live, instead of catching it on iTunes U or downloading older videos for later viewing? Just curious if there's something I'm missing something that adds value one way or another.
As I understand it, it's not only about watching the lectures but actually taking the course (with assignments and all). For that to work, it makes sense to be on the same schedule as the "real" students.
That said, I'd quite like to just watch the vids sometime later. Anyone know if / where they will be available?
Cool that so many things like this are popping up lately:
Tom Mitchell at CMU putting up all of his lectures/material, the Stanford/coursera ML course, now this CalTech course... are there others?
was mostly thinking of courses that had lectures/vids up, but that also reminds me of this berkeley course i came across on HN a couple of weeks ago:
http://alex.smola.org/teaching/berkeley2012/
