Here are my notes (on github). Nowhere near as polished as this version, and probably reveals more about my process of understanding than machine learning itself, but if we're sharing: https://github.com/mechamoth/ml-class/blob/master/ML_Notes.o...

Thanks for the great resource. I wish there was a way every student could share notes with everyone.

My friend and I have been thinking about this for a while.

Q.How would people share notes? A) Could be live sharing in class (etherpad for ex: http://piratepad.net/sj8l1FIUIK ), then contents of pad get transferred to a wiki and people are free to edit.

Q2.Why would people put in the effort to type-in their notes? A2) Therein lies the trouble...

If you can think of a business model, or an incentives structure that would make this a open collaborative project, do pm me. Note: It would be against my values to put ads all over site to monetize.

In statistics in college, a I and a bunch of my friends (CS students) took cooperative notes in Google Wave (which, by the way, I still maintain was amazing). Generally, a friend would be taking notes as quickly as possible, and I would follow him and replace his ad hoc notation with the correct unicode mathematical symbols to make everything look really nice. But then we found all the homework solutions and sample tests (by looking at the format of the URLs of the previous ones he gave us).

Great to hear I'm not the only one enamored with Google Wave.

When I was in Germany, a group of students (who wanted to do it) were given the task to take notes in latex. Then every week they compiled the pdf and was put on the website of the faculty.

To write the latex file, they worked together (sharing the same file, same idea as google docs).

And at the end, they put the effort only because they thought it was nice to have all the notes in pdf, commitment with the class, and because it was a good idea to share it with the other students. I still have all my mathematic classes pdf's.

I believe this is commonly called "scribing". A lot of professors over the world follow the same for their courses.

A1 : Live sharing may take some time to achieve. Even if we reach to a stage of offline sharing, it should be a very decent utility.

A2 : Incentive model may always not necessarily involve monetary benefit.

PS : Check mail.

I used this site a few times in college - http://www.notehall.com/ . The downside is you have to pay for notes, study guides etc. but they let you preview the notes and then leave reviews after you've purchased.

Just something that caught my eye: In the introduction, you have written "clarification problem", which of course should be "classification problem".

Cheers - updated!

Thanks for this, especially for annotating the lectures. It's much easier to skim/review this way. I was rewatching the computer vision pipeline lectures over the weekend when it occurred to me that they might not be available when the new class starts.

Thank you good sir, wish more people shared their quality notes like you!

Wow! What a great resource, thanks!

What math preliminaries are necessary to understand machine learning? I'm taking a basic linear algebra and probability course in the Spring...I'm wondering if that would be enough.

Complement this course with lessons from khanacademy as soon as you stumble upon a concept either you do not know or want to know a bit more than presented by the class. The class is pretty self contained however I found this practice very helpful.

Khanacademy classes are mostly 10 min video lectures, so you can jump to any concept and learn very quickly.

One can not eat an elephant at once, but if you slice it and eat a bit everyday you'll go a long way in short period of time.

www.khanacademy.org/#browse

In this specific course, not much. I personally had no maths education beyond year 12, and only what I'd picked up along the way while reading around the subject. All the maths you'll need to know is explained thoroughly enough in the lectures. I think if you already know linear algebra, you'll have a head start - I didn't know it, and it took a bit of work to get up to speed.

There is a lot of discussion of derivatives and partial derivatives in the course, however it's entirely possible to complete the course and feel comfortable with the material without understanding derivatives. At least, it was for me =)

As for machine learning more generally, I'm certain a greater understanding of the maths behind what's going on would make you a better practitioner. I'm sure there are plenty of people currently making their living applying ML techniques without being maths experts, but I'm also sure the leaders in the ML field are maths experts.

Why not a git repo on Github? So people can contribute.

Mainly because the HTML Evernote pumps out is pretty disgusting - it works in terms of making the markup look like it did in Evernote, but it's not structured and does some pretty interesting things.

I'm hoping to work on cleaning it up over the next few weeks, at which point a repo might be a good idea. My only slight concern is that if every man and his dog contributes we may end up with a really thorough, in depth overview of the topics, which in one sense would be brilliant, but on the other hand would make it far more than just a simple reflection of the course.

However, that's hardly reason not to do it, and there can always be one copy in it's current, terse format, so when I get the HTML sorted I'll do that and see what happens....