I get easily excited about education-related topics so I may be over-reacting, but I think these classes will jump-start an educational revolution, and that people will start to fully appreciate just how inefficient traditional teaching methods are.
Some people like to say that this is nothing new because video lectures were posted on the internet for several years now (for example MIT Open Courseware etc.), but I think this misses the point entirely. There is a huge difference between low-quality video/audio recording of a prof mumbling for an hour and post-processed, perfected snippets of videos presented in a coherent fashion, and most importantly with supplementary materials that encourage people to actually apply their knowledge and get feedback. In addition, the fact that many people take the class at the same time also enhances the experience for everyone, and we've seen study groups form everywhere around internet.
Full disclosure, by the way, I'm a CS PhD student at Stanford and I am a (voluntary) co-creator of the programming assignments for the current ML class. It is a lot of work, but the way I see it, we only have to put great assignments together a single time, and thousands of people can enjoy them and benefit from them for years and years to come. That is what I call time well spent.
I hope all these classes go well, and I'm looking forward to telling my kids about what education used to be like in the old days. I have a feeling that they'll find it hard to believe me.
" I'm a CS PhD student at Stanford and I am a (voluntary) co-creator of the programming assignments for the current ML class. It is a lot of work, but the way I see it, we only have to put great assignments together a single time, and thousands of people can enjoy them and benefit from them for years and years to come. That is what I call time well spent."
As a consumer of Stanford's online classes(though not the ML class. I'm waiting for the CS 229 - vs the CS 229A - version), let me take the opportunity to thank you. Your efforts are totally appreciated. You are right, this is revolutionary. Glad to see that Stanford is keeping up (and building!) momentum rather than this being a one off effort.
thanks, I should add (and I hope this is somewhat obvious) that there is a team of about 7 or 8 of us in total. A lot of work goes into site/technical/video processing etc, and then there are 2.5 of us making the assignments.
The real hero behind the assignments is Jiquan Ngiam (http://cs.stanford.edu/~jngiam/) who is in charge and does most of the work. He is awesome, brilliant, very hard working, and I thoroughly enjoy hacking with him on the ML class until late AM.
Me and my wife have been following Professor Ng's ML class and one of the highest points for us has been working through and trying to understand the programming exercices.
Our personal thanks to you and the other "1.5 people" working on that. ;)
Thanks for that. I suspect Prof Ng will get a lot of praise for the class but I suspect some of it is intended for the people behind the scenes (not always easy to find out who they are).
Thanks to all you guys! The programming assignments are brilliant. All my (non-technical) friends and family I've been telling about the class think the whole thing sounds really cool.
Really great stuff. Thank you and your team for your time and efforts on this. I learned a lot already from the ML class and can't wait to take the HCI class.
The homework and in-video questions are what keep me involved. I have such a busy schedule that its difficult to maintain the discipline to simply keep up if there were only videos. However, having external deadlines and practical things have encouraged me to make the time (and I'm very happy about it).
I agree that this is potentially revolutionary. Before these courses, I couldn't have imagined doing a 'distance-learning' course. I suspect there are many other folks that feel the same.
Also, thank you for your work on the ML programming exercises! I've found them fantastic in getting my head around how to practically 'encode' the things from the videos. Much appreciated.
The time management part is difficult. I tried to keep up with ML but because I kept thinking, I'll do it over the weekend, I got a couple of weeks behind and ended up dropping. I'd love to join a study group in conjunction with the class - then I have the social pressure to keep up (but not necessarily share grades..)
I think independent study groups are the big win from this. This isn't the first time high quality videos have been online (the full "regular" ML class can be viewed on youtube), but because it's more "real-time" it allows for groups of people to work through the material together. It would be significantly more difficult to co-ordinate that with the MIT videos that you can watch whenever you'd like.
Both are great resources, and I don't know that the Stanford classes will be able to sustain their ability to get people to create their own study groups. Our study group here in Cleveland Ohio gets 5 - 10 people each week, but we're drawing from areas almost an hour away. Future ML and AI classes will have a tougher time having that same draw, as those most interested will have already gone through the material and would be much less likely to drive an hour each week for a study group.
Survey classes like the ML and AI class work very well with this format. I'm looking forward to seeing how it fits with more specific classes. There are times in the AI class that are a bit "hand-wavy", which is OK for an over-view/survey class. I'd find that a bit more annoying on something like the upcoming natural language processing class.
"Future ML and AI classes will have a tougher time having that same draw"
Really? Or are you pioneers? I know reading your comment just greatly increased the probability that I will take another class (currently in db-class) and seek out a study group next time.
I'm a big fan of the study group setting for these classes. It's been a great way to find and work with like-minded people. I'd encourage you to find one for future classes, but I'm not sure how sustainable they are.
The AI and ML classes drew a lot of attention to start with (over 100k signups). Follow up or repeat classes will obviously draw less interest. I'm interested in seeing how that translates to viable study group sizes. The good news is that the minimum viable size of a study group is fairly small.
thanks, I agree with you-- there is something psychologically interesting about it that I can't quite understand, but it certainly seems like this format better maintains student motivation. It's a step in the right direction at least... just the first iteration of a gradient descent ahead of us ;)
Also, good point about the programming exercises. Especially with topics like this I think it is easy to fall into trap of thinking you "got it" if you just watch the videos. Programming exercises force you to think it through on a whole different level and, I think, lead to better understanding and retention.
> there is something psychologically interesting about it that I can't quite understand
I swear, it's an extra pleasure hearing a song I like on the radio (over just playing it myself), knowing that a lot of other people are listening to it too. Maybe that's a new form of 'social computing' to be exploited :).
I've been watching Susskind's physics lectures on youtube, which are also from Stanford (continuing education). They're just video of the lectures, no homework or quizzes.
They're great stuff and he's really gotten me to understand a lot of the "why" of physics rather than just the "how".
Having said that, I'd love to have homework along with them, even just suggested problems with an answer key.
I have a masters in CS, and I am taking the online ML class. I highly appreciate how well organized the class is. Thankyou.
I have a question though, is there a way these classes can be made to be taken anytime a student wants?
The problem I am facing is that since I am also working fulltime, I just in time manage to submit the homework, and as a result, I can not take more than one classes at the same time, like the DB and AI classes in this case.
We'll have to see what happens to the evaluation systems once the term ends, but at the very least you can sign up for all the classes and download the videos.
More importantly, for evaluation purposes, the quizzes and exercises can be submitted late, though penalized in points. Since what you get out of the course for having a high score (a certificate/letter of completion?) is worth about as much as toilet paper, you can still do the exercises and be evaluated on them even weeks later. We'll have to see if they keep the system running past the end of the semester - probably won't happen with ML, but maybe DB will stick around a bit.
AI, well, that doesn't have any homework.
So, next semester, sign up for all the courses, stick with all of them past the introduction week, pick one or you you'll focus on, and dabble in the rest.
If you don't need concurrent learning to be motivated you can watch videos later. I wish there would be a way to be evaluated without any after-deadline penalty.
I'm taking the ml class right now, thank you! Everything you say is absolutely true, enrolling in this class vs watching some videos is like night and day:
- a large number of other students who are at the same cohort helping each other
- lecture materials
- programming assignments
- comprehension questions
I don't feel quite like I am taking a real masters level computer science class mainly because the assignments are easier (you guys set up a lot of the boiler plate for us and we implement a few core algorithms). That said, the assignments have been a big part of me retaining the concepts and since they are not ball breakers, I can actually keep up with 30-60 minutes a day, with perhaps a couple hours one day on the weekend.
Professor NG is also an amazing teacher. I'm cautiously optimistic that the other teachers will be as good.
I definitely agree with you on the "huge difference between low-quality video/audio recording of a prof mumbling for an hour and post-processed, perfected snippets of videos presented in a coherent fashion, and most importantly with supplementary materials that encourage people to actually apply their knowledge and get feedback" part.
I started the site noexcuselist.com as a page to direct people to the best places to learn on the web. In doing so, I had to go through tons of web pages claiming that they taught things for free. The sites that I was really excited for like the Open Courseware sites were a bit of a disappointment for me. It'd be pretty hard for someone to learn a entire topic using it due to the incomplete lectures, some classes that weren't available, and the lack of lecture notes and homework that went with it.
I'm pretty excited to see the development of this one that you're working on though. Good luck and keep us posted!
Jason Kuen Wen Yong's comment mentions (and I concur) that there are additional topics in the Jan 2012 course than the current one like deep learning. Any clarification on this?
Seconded. For someone who is being newly introduced to ML concepts, applying them right away can be daunting. The exercises do just enough hand-holding to make sure they don't get lost, but give enough leeway so that they know what kind of mistakes they are making.
And the submission process is just mind-numbingly simple (which is good!).
I largely agree with you, but, I feel that the in class presentations, when recorded well, are more engaging.
Personal opinion being what it is for me I find it far easier to listen to Professor Sahami's recorded lessons (CS 106A) for an hour than I do Prof. Widom's into to DB classes. This is not intended as a slight to any person at Stanford. The real difference comes down to watching someone engage an audience and some one speaking to a camera.
Strictly my opinion based on a very small sample pool.
I second this, I am in the db-class and I can't help but think it would be more fun with an interactive class. Widom is a great teacher, but it's weird to just watch her in that room by herself talking about database stuff.
I'm taking the ML class and I love it! I agree with everything you said above, and I'd love to see a more advance ML class in the future. Keep up the good work!
Thank you very much for your efforts, I've been enjoying ML class so much. The only thing I disagree with, is that the courses are time-bound. I have a day time job, and I'm only able to take 1 course. However I don't see any reasons why you shouldn't allow people getting through the courses outside of the time frame. This would be more convenient for a good part of the target audience, because it's the people who actually don't have enough time to enroll in the real university course.
E-learning has the potential to: decrease the costs of getting an education, create the potential for "mass customization" of education, reduce credentialism in society, make learning an end in itself for many people, force universities to become less complacent, and probably many other effects I am overlooking right now.
Starting an e-learning startup is something I REALLY want to do. I love learning and I love startups.
I hope I can make that happen someday (nudges everyone with similar dreams).
Karpathy: Do you know if prof. Ng will cover the issue of sparse data in his lectures?
When the number of dimensions is much greater than the number of samples and most of your cells in a matrix are equal to zero then most of the ML algorithms don't behave too well. It's very common problem in NLP to have sparse matrices.
Any chance (some of) the code for the platform could be shared? I particularly love the exercises (review questions) and the way some questions are embedded in the videos. Thank you and your teammates for some great work.
Thank you. There is attention to detail here. It actually amused me when I watched Andrew Ng's videos at 1.2 or 1.5 speed that his voice didn't become squeaky because it is frequency modulated.
Thank you for your work. I am in both the AI and the ML classes, and having some work that we actually do stuff, is a great learning aid and I wish the AI course had something similar too.
Just want to say "Thank you!" for all the hard work you and your colleagues put into ML class. This is the best learning experience I've had in my entire life.
Some people like to say that this is nothing new because video lectures were posted on the internet for several years now (for example MIT Open Courseware etc.), but I think this misses the point entirely. There is a huge difference between low-quality video/audio recording of a prof mumbling for an hour and post-processed, perfected snippets of videos presented in a coherent fashion, and most importantly with supplementary materials that encourage people to actually apply their knowledge and get feedback. In addition, the fact that many people take the class at the same time also enhances the experience for everyone, and we've seen study groups form everywhere around internet.
Full disclosure, by the way, I'm a CS PhD student at Stanford and I am a (voluntary) co-creator of the programming assignments for the current ML class. It is a lot of work, but the way I see it, we only have to put great assignments together a single time, and thousands of people can enjoy them and benefit from them for years and years to come. That is what I call time well spent.
I hope all these classes go well, and I'm looking forward to telling my kids about what education used to be like in the old days. I have a feeling that they'll find it hard to believe me.