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.
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.
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 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.
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!
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.
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.
I, for one, am really hoping this is a start of a trend - where coursework, even if just in a prerecorded format, is available to all, with the tuition going towards the rich in-person experience and grading/certification for the student, much like the way the primary tech conferences have been trending for free content for al.
MIT OCW requires a far higher level of intrinsic motivation. Stanford has almost nailed it. While online learning has taken Space out of the equation, Time, it seems, is still a big variable. Having people do the course at the same time with deadlines is working. They just need to work through the technical glitches, which shouldn't be that hard.
This is fantastic. I would also like to do some course in mathematics (under graduate level calculus, discrete mathematics ) to improve my skills. Are there any good places/resources where I can learn these things (video lectures with quizzes in between will be a nice choice). I am also ready to take a paid certified course if some reputed college is offering them online.
Although I had to drop the previous classes due to time, this looks promising. Particularly with the unified style; that is the one thing that seemed to hurt this years effort - The most widely advertised course (AI) had the worst layout and 'features'.
Both ML and DB classes are better executed than AI class. AI class is backed with a broken platform by a small startup, whereas ML class was run by code developed in Stanford. AI class also had jarring video editing which totally broke the rhythm. Both ML class and e.g. Khan Academy has this natural flow that is pleasure to follow.
To me it has been surprising how big difference in experience the technical execution between classes made. Both Norvig and Thrun are clearly good teachers, but AI class failed because of the poor technical implementation.
The video editing is also fine. If such minor issues break your rhythm, you might need to rethink your approach to the lessons.
Video editing in AI class is against the knowledge that we have learned in 100+ years of movie editing. Cutting 1-2 seconds but keeping the same scene gives a jarring experience.
They cut out a few seconds so that you don't need to watch Norvig or Thrun to finish writing of a word or sentence, but that's saving time in the wrong place in my opinion. If you watch Khan Academy or ML class, those 1-2 seconds give a natural pause, that you can use to think about the subject.
I've nothing against the pen and paper approach, although occlusion is sometimes a problem. Based on my experience with Khan, AI and ML classes drawing by hand is the essential thing, is it less important if it's done electronically or by pen and paper.
Anyway, my main point is that execution of e.g. Khan Academy and ML class are better than AI class, and I'd personally prefer that inspiration for new online education courses is taken from them instead of AI class approach.
Try watching the ML or DB lectures in '1.5x' mode (it speeds the video up). Because of the faster pace, it is a really good counter to my ADD tendencies. Also it is really nice when you are trying to cram a few lectures into a short time period.
I wouldn't call the ai-class.org platform broken. It's just not done yet.
The AI class is like a beta test and there are plenty of bugs and problems. The content is still good, but the programming problems and automated grader that would be necessary for students to really absorb the advanced material doesn't seem to be on their radar. The result is a certain superficiality.
The ML class is much more developed. In the first few weeks the contrast of the finished, professional seeming ML class with the rough and hacky AI class was extreme but AI is catching up.
I'm looking forward to improvements in both platforms and new material in the spring. Other competitors like Kahnacademy are exciting, too.
I'd like to see more study groups around the world building this into a movement with broadening impact. Neither class had direct support for forming study groups. If you have to go on Reddit to get together with people, you're excluding 50-90% of the people who would use a simple no-friction geo-aware meetup-style tool if it were on the site's front page and main menu. Since study groups have a minimum size threshold to remain viable, that eliminates a magnified portion of potential participants.
So I'm pleased and still hope that this could be a mass-movement that leverages technology to change the way education is done around the world. Competing platforms is a better bet than one great platform for perfecting hard to build but needed features like labs, programming assignments, study group support, office hours and teachers' assistants, and homework help and hints.
I started out spending closer to 10 hours a week earlier on in the AI course. We initially had a local meetup and I spent more time reading the textbook for the first few units. Had to scale it back to 5 or so hours, which is good enough to get through it, but not enough to spend extra time discussing with others, reading supplementary materials, and coding.
I've put about 2-4 hours a week into the ML class. However, I know that math (basic linear algebra and calculus so far) and have done scientific computing before so I can churn the exercises out pretty quick. The exercises are the best part. It's nice having a set of problems to work on instead of just watching an hours worth of videos each week.
For the ongoing courses it will be available, so I guess it will be the same for these also.The thing you miss out on taking them later is the structure of the class with deadlines (very good incentive to really sit down and learn!) and personal letter from the teachers
Do you have a reference for that about the ongoing courses? I tried to find something about it a while ago and couldn't. It would be awesome if they plan to just leave the whole web app with quizzes etc up and open for people to use at their own pace.
Q: Will the materials still be available after the course is over?
At the very least, all of the materials in the OpenClassroom repository will be available, including videos with embedded quizzes, lecture notes, course materials, software guides, and extensive do-it-yourself exercises with solutions.
I was actually wondering about the AI class... it's not listed on the Open Classroom site. It's really weird that they are not under the same umbrella but I guess the AI class is managed by that start up or something? Thanks anyway.
For me, as a brazilian computer engineering student, I think these classes are amazing. Although I study in one of the best computer schools in brazil ( www.cin.ufpe.br), my classes tend to be bad and boring. My teachers have Phd's and all those letters, but cant teach in a good and engaging way.
And my classes have old subjects, because they made the curriculum 10 years ago.
Thank you so much, Stanford, the teachers for the modern and brave choice to teach people all around the world, thanks for all the students engaged in making the classes available for everybody. and thanks hn buddies for always giving the good news.
Arghhhh. I find these incredibly frustrating. I am writing a senior honors thesis for my university, and wanted to take as few classes as possible this year to focus on my thesis work. The ML class has distracted me this semester, and it looks like things are going to get worse next semester.
On a serious note, does anyone know if there are plans to continue these courses next year? I suppose it will really depend on how well each class goes, but I mostly feel pressured to take these because I'm afraid I'll miss them.
This is very exciting news, very much an autodidact's dream.
I've been enjoying crawling my way through the great ML classes by Prof Andrew Ng, and had been wondering if by any luck other classes would be provided for future semesters. Seeing this just makes me really happy, and thankful to Stanford. Not only is it the future of education but also gives countless people around the globe a chance to learn topics they may otherwise have never had access to (I am one in this category).
Unfortunately, Blackboards' patents preclude this from becoming a reality. For the next 15 years, Blackboard has an exclusive right to practice this "virtual classroom" business method.
Their patent pledge promises that they will not assert their rights against any open-source or home-grown initiative, which is why Stanford et al. are able to get away with hosting their own content in this format.
It's a harsh reality that by itself makes the case for patent reform.
The lectures are broken up into segments (which you might find convenient or annoying). There's not much more than that, unfortunately. If you want exercises, his problem sets are mostly from Kleinberg & Tardos, sometimes with problems from CLRS.
One of the great hidden benefits of those classes is to taste the flavor of the classes, should they want to apply to the university. I would particularly interested in the difference between Berkeley classes (SaaS) and Stanford classes (CS101 or ML)
Join all the ones you're interested in then drop the least interesting classes. I joined the ML and AI class this year then focused on the ML class because the octave exercises are a great way to learn.
The probabilistic graphical models looks like it may have a similar format as the introductory video mentions automatically graded programming exercises.
That would be great. I was disappointed when the AI class didn't have programming exercises. I feel like these were the part that really enforced the learning for me.
I think another part that worked well was that getting 100% for each exercise is readily achievable. I want to finish each exercise perfectly so I don't break my 100% record for the course. I think this works well for motivation.
As it names suggests, HCI is a field that attempts to understand the way we interact with machines. The idea is that by understanding this relationship we should be able to build better designed software/hardware (or more usable, as it's known in the field). It's a merging field involving mainly psychology, design and computer science. As for how it applies to the real world, either UX design or usability engineering are probably the most commonly found examples out there.