MOOC plans


(Jeremy Howard (Admin)) #1

Any ideas for, or questions about, or comments on the MOOC that we’ll be creating from this course? Any thoughts about how you might be interested in getting involved? How can we make it as successful as possible?

All input most welcome! :slight_smile:


#2

Will you be using a MOOC platform like Coursera or EdX? I wouldn’t mind taking on tasks to help automate the setup of deep learning projects, and contributing the code back to open source, if possible.


(Jeremy Howard (Admin)) #3

I’m not planning to use a platform - I was planning to host the videos on youtube and use the wiki and notebooks for notes and problems. I suspect using a platform would be a lot of work, and probably wouldn’t allow us to use our existing videos.


(lin.crampton) #4

Jeremy, I’m going to be reviewing all of the class videos over the next week or so. Do you want me to do a transcript of any of these videos while I’m watching? if you use these videos as the basis for a MOOC, you will probably need transcripts.


(Jeremy Howard (Admin)) #5

@lin.crampton that’s extraordinarily generous! I was planning to use speech recognition. Although Dragon is still not close to start of the art, so there will be errors - and more importantly, it doesn’t do punctuation (unless that’s narrated specifically).

If you feel like creating transcripts is something that wouldn’t detract too much from your watching, then yes, it would certainly be a wonderful thing to have - but I wouldn’t want you to do anything that distracts you from picking up the material.


(learner) #6

I am trying to maintain a running version of your notebooks, if that is of interest I can put it in a public repo for your consideration.

As for ideas, one problem with moocs is that they have a very high dropout rate (no pun intended). One way to improve student engagement in the mooc would be to start a development project of a chat bot, who can answer questions about deep learning in general and ask questions specific to the material of the course (and I’m counting on learning enough NLP in the rest of this course to get it working.)

Beyond the uncertain degree of success that the bot may have, it could work as a ‘capstone project’ for the mooc and it would keep students motivated to complete the mooc because this would allow them to contribute in a real AI project. I would gladly volunteer some weekly time to help get a project like this off the ground if there is enough interest.


(Jeremy Howard (Admin)) #7

Thanks for your thoughts and offers! What do you mean by maintaining a “running version” of the notebooks? - Do you mean having them hosted somewhere so that people can login and play with them? Or do you mean static versions like on github?

A chat bot might be a better project for part 2 of the course, since we’re not really getting to Q&A techniques in this part (largely because they’re still not very practical IMHO).


(chris) #8

I was just thinking that the notebooks and supporting files belong on github.


(Jeremy Howard (Admin)) #9

Yup they’ll move there as we wrap up part I. Just wanting to avoid having to learn any more tech bits than necessary for this first part of the course.


(chris) #10


(Jeremy Howard (Admin)) #11

Yup - that’s just the slide I’m planning to use when I introduce it! :smiley:


(yad.faeq) #12

Here are some notes that might be helpful from my previous experiences at building online school / course content in some shape or form of MOOC, DU-Online & The Bitcoin Course:

  • Short videos: After talking with Udemy and Udacity, we were told repeatedly that short videos work better in general, due to:

  • Short attention span and distraction on Internet or phone.

  • Allowing learners to re-watch a specific topic over and over without watching a whole 15 minutes.

  • Notes or Transcripts: Having course notes allow the student stress less about taking notes and rather focus on the material being taught, which is a huge drawback for online students when the work becomes too much.

  • Supplementary Material: Providing not directly related material can be things like,

  • How to do Matrix Multiplications.

  • Parsing text with NLTK, Spacy in Python.
    This allows the student to stop looking around and getting distracted to learn about a topic that is partially relevant to the course. We already have some version of this through Blog post listings and such.

  • Another great approach to excel at this point is Videolectures.net, for example we can take this talk by Hinton, the Slide is providing side-by-side when the talk continues and the speaker changes slides.

  • Closed Captions: Aside from providing course material to people with hearing difficulties. CCs are used by people that play videos on fast-pace 2x speed, or just when not might be paying attention to the audio.

  • Access: There are countries with huge user-base for online learning that don’t have access to either:

  • Hosts that the videos are offered on, such as YouTube and Vimeo.

  • Quality of videos for Internet Speed, in places that can’t get fast internet access.

  • Platforms that are more than just Web, as in availability through mobile phone version, which refers back to the previous two points I guess.

Lastly,

  • Motive: Providing motivation to students to finish a course increases not just the rate of course being finished, but an over-all satisfaction from those that drop-out as well. This was something suggested to my team after meeting Udacity’s team. For example:
  • Half-back, for the course if a student successfully finishes everything.
  • Certificates of work that have been completed (actually completed though, not copy paste). Udacity currently calls this Nano-Degree
  • Project-based courses. For example, we tailored talks from founders to allow a student learn how to make a pitch deck. While Udacity currently says, let’s build an Autonomous vehicle. The notion of providing an end goal project gets the student on to a journey, helping them through struggle, googling, debugging and all that.

One last piece, having built an actual online school product. The technology doesn’t really matter to make the best out of the content. Useful resource and videos will get passed around, the rest of it is like fine-tuning a model towards specific target audiences.


(Jeremy Howard (Admin)) #13

Thanks @yad.faeq that’s very thorough!

I should set your expectations however - the goal is to get something up ASAP after the course is finished, so that people have time to study it prior to part 2 of the course next year. So some of the things you suggest that are more time-intensive are unlikely to happen in this time-frame!

Personally, I prefer the small thumbnail video so that the student can focus on the slides - I’m not thrilled by the side-by-side presentation. With most monitors, the side-by-side version ends up with really small slides!

For access - where are good places to host that are more accessible? I’m only aware of China as being a large market that is blocked from youtube - are there others? I think it would be a great idea to have a very low bandwidth version of the video to help those in areas with crappy internet (like Australia :wink: )


(yad.faeq) #14

No worries, I had those notes in general about building MOOC courses and thought one of them might be helpful.

Indeed, Youtube is best out there in terms of the offerings such as videos to be encoded to different qualities.

To my knowledge China is the only permanent one, though Wikis say this:
Access to Youtube has been at one time or another blocked in Brazil, Turkey, Germany, Libya, Thailand, Turkmenistan, China, North Korea, Iran, and Pakistan.

Not all of them are permanent, and I think people find their way usually.


(anamariapopescug) #15

i’ll be reviewing the class videos starting next week and if you’re interested i am happy to send feedback as to what was more unclear, etc. (especially for the later NLP parts, but not only).


(Jeremy Howard (Admin)) #16

That would be very helpful - posting your thoughts in a dedicated thread on the forum would be a great way to keep these all in one place. Thanks for the offer!


(sravya8) #17

When is the MOOC going to start by the way?

I strongly +1 for use of github. Otherwise, it is very easy to miss changed files. IMO, it would be net positive, I assume a good portion of people would be familiar with git. If not they just need to learn using three commands: clone, fetch and pull. Maybe we can add aliases for these as well to make it slightly easier for GitHub newbies.

In the MOOC, do we plan to release all class material and videos at once or do we have a dedicated class schedule? If the latter, I think it would be useful releasing the notebooks prior to the class along with class agenda. So that students can have a chance to do some reading (especially prerequisites) before the class ready to grasp the material.

Besides that, I would like to be involved during the MOOC and help out in areas of my expertise. Although not sure what is the best place for me to contribute yet :slight_smile: Some broad categories where I think students will need help:

  1. Setup issues
  2. Understanding the material
  3. Further reading and learning
  4. Kaggle competitions
  5. Capstone projects
  6. Jobs and careers

I think we have reasonable to good content in 1-4 through wiki and forum and can build on it during the MOOC. Would it be useful to start collecting data for 5 and 6?

For 2: I am a visual person and for me, personally a short video (Khan academy style) summarizing the class content would be extremely useful. We can either link if there is already one which exists, if not we can try making one?

For 3: There is just too much reading content out there, it would help if we had something like : [Title, votes, time to read, summary] so that people can choose their reading material based on their time availability and interest areas. Not sure if there is a tool/framework which we can leverage here. Ideas welcome! Also, a weekly ( or bi-weekly) paper reading group might be useful, as it does seem like there is tremendous value in learning the skills to keep ourselves up to date even after the class. Also, we should probably minimize extra reading during the class, as I feel the class content itself is sufficient to keep students busy during the week. So paper reading group would be ideal for after the course completes.

Apart from that, we should probably also think about setting up analytics for the MOOC if we have not already. Will we know how many people watched what portions of the videos and when did people start dropping off? And analytics on how many students read notes, how many submitted on Kaggle and so on?


(Jeremy Howard (Admin)) #18

I don’t know - but it’ll have to be very soon, for students to be able to get through the material prior to part II! Maybe 3 weeks or so?

Agreed.

I’m not sure. Most MOOCs release gradually, but I’m not sure why - it seems like more work for us to do it this way. I agree that providing all the materials for a class at the start is the best way to go.

Wonderful! Clearly we’re going to need lots of help answering student questions on the forum, as well as improving the wiki. I think there’s a lot of room to improve the wiki - on the whole, frequently asked questions are not having their answers transferred to the wiki as often as would be ideal, which means that the main source of answers right now is still the forums, rather than the wiki. I don’t think that’s ideal.

That’s an interesting idea. So you’re thinking that this would be something they’d watch prior to the main lesson video, giving an overview of what is going to be covered? What would be the main purpose of that?

I don’t know about the votes idea, but the other bits we can easily do on a wiki page. A reading group would be a fun way to keep the community engaged!

I’m assuming that the videos will remain hosted on youtube. Does youtube provide that info?

I don’t see an easy way to track reading of competition submission behaviour.


(sravya8) #19

Completely agree! Will start contributing to the wiki in the following weeks.

I am thinking in terms of a video version of each lecture outline (3-4 minutes) or rather "A 10-15 minute outline of part 1 or a couple of videos for major concepts/ problem areas. I kind of feel I am understanding bits and pieces but it would take a while (go over the content again) to get a good grasp of everything that was taught. I will be interested in doing this along with others who might be interested. Main purpose is a short overview video which covers main take away points and has pointers to individual lectures when necessary.

Cool! Will start another forum thread to start a discussion around it.

Just checked youtube analytics and it does have Audience retention metrics, here is a sample of one of my videos

Although I would imagine MOOC websites like Coursera, Udacity might have better analytics tailored for online courses.


(Jeremy Howard (Admin)) #20

These both sound valuable. Perhaps you could seed a forum thread where we could hear from you and others what you feel are the key points you’d like to see covered in both the high level outline of part 1, and the lecture outlines? I’d be happy to then create those videos.