Machine Learning (Theory)


To Vidoelecture or not

Tags: Conferences,Machine Learning jl@ 1:21 pm

(update: cross-posted on CACM)

For the first time in several years, ICML 2010 did not have videolectures attending. Luckily, the tutorial on exploration and learning which Alina and I put together can be viewed, since we also presented at KDD 2010, which included videolecture support.

ICML didn’t cover the cost of a videolecture, because PASCAL didn’t provide a grant for it this year. On the other hand, KDD covered it out of registration costs. The cost of videolectures isn’t cheap. For a workshop the baseline quote we have is 270 euro per hour, plus a similar cost for the cameraman’s travel and accomodation. This can be reduced substantially by having a volunteer with a camera handle the cameraman duties, uploading the video and slides to be processed for a quoted 216 euro per hour.

Youtube is the most predominant free video site with a cost of $0, but it turns out to be a poor alternative. 15 minute upload limits do not match typical talk lengths. Videolectures also have side-by-side synchronized slides & video which allows quick navigation of the videostream and acceptable resolution of typical talk slides. Overall, these benefits are substantial enough that youtube is not presently a serious alternative.

So, if we can’t avoid paying the cost, is it worthwhile? One way to judge this is by comparing how much authors currently spend traveling to a conference and presenting research vs. the size of the audience. In general, costs vary wildly, but for a typical academic international conference, airfare, hotel, and registration are commonly at least $1000 even after scrimping. The sizes of audiences also varies substantially, but something in the 30-100 range is a typical average. For KDD 2010, the average number of views per presentation is 14.6, but this is misleadingly low, as KDD presentations were just put up. A better number is for KDD 2009, where the average view number is presently 74.2. This number is representative with ICML 2009 presently averaging 115.8. We can argue about the relative merits of online vs. in-person viewing, but the order of their value is at least unclear, since in an online system people specifically seek out lectures to view while at the conference itself people are often opportunistic viewers. Valuing these equally, we see that videolectures increases the size of the audience, and (hence) the value to authors by perhaps a factor of 2 for a cost around 1/3 of current presentation costs.

This conclusion is conservative, because a videolecture is almost surely viewed over more than a year, cost of conference attendance are often higher, and the cost in terms of a presenter’s time is not accounted for. Overall, videolecture coverage seems quite worthwhile. Since authors also typically are the attendees of a conference, increasing the registration fees to cover the cost of videolectures seems reasonable. A videolecture is simply a new publishing format.

We can hope that the price will drop over time, as it’s not clear to me that the 216 euros/hour reflects the real costs of Some competition of a similar quality would be the surest way to do that. But in the near future, there are two categories of conferences—those that judge the value of their content above 216 euros/hour, and those that do not. Whether or not a conference has videolecture support substantially impacts its desirability as a place to send papers.

31 Comments to “To Vidoelecture or not”
  1. Lev Reyzin says:

    Interesting interpretation. Another way to think about it is that the primary beneficiaries of a video lecture are those people _not_ attending the conference, as the attendees can see the talks live. Under that view, increasing the registration cost for attendees to support non-attendees might not be a good idea. And putting the cost on the authors makes them pay more to publish, which is also not my favorite model.

    • Actually, after attending NIPS 2009 I’ve watched again on videolectures nearly all lectures I did attend at the conference, to refresh what I had seen and/or look for references or clarifications to some marginal points.

      Also, in multi-track conferences (like the NIPS workshops) it is useful to be able to choose with less guilt which talks one should not attend, if one can view them online later.

  2. I know that some of the other conferences like USENIX for instance don’t use videolectures, but get individuals/organizations who only provide the recording and post-processing services for a cheaper cost. The hosting part doesn’t seem like that much of an issue. All our conferences have websites that stay up forever, and can host any video content from the conference. Might be a much cheaper long term alternative to videconference.

  3. Brian Tanner says:

    I don’t know if I agree that most of these websites will stay up forever, and there is a great value in having the content all in one place, so similar topics can easily be connected together. I think the concept is great and having these excellent talks by these top minds from our fields is extremely worthwhile.

    However, I always have wondered at the cost of capturing/hosting these via videolectures. I expected that it would be a very pricey endeavour, which apparently it is. It seems perhaps like an opportunity for a more competitively priced alternative to enter the market.

    I wonder if, after paying these high costs, the conferences have access to the original data/videos/presentations. Would it be possible to migrate/duplicate existing lectures to a different service if such a service existed…

  4. Tomas Singliar says:

    We are working hard to find affordable ways of having high-quality talk videos available from ICML 2011.

  5. Kevin Canini says:

    I definitely saw video cameras recording the speakers at ICML 2010. If that wasn’t for videolectures, does anyone know where the recordings went?

  6. Weiwei says:

    “A videolecture is simply a new publishing format.”


    Video + slides convey a more comprehensive information and reduce the ambiguity in the paper. It is definitely a great complement. Most people I talked with during ICML this year felt it was a pity, not to mention the people who were not there.

    But 216 € per hour? That is too much.

  7. Shubhendu T says:

    With respect to your observations about youtube.
    I am wondering how much does it cost for the IITs (, MIT, UCs or Stanford etc to be allowed to upload videos > 1 hour.

    But you are right, the best part about videolectures is indeed being able to see the slides on one half.

    • Lars Pellarin says:

      I never had much use of the fancy parallel video lecture slides. If i need slides, i usually download them (if available) as an additional file to refer to.
      In this regard, i think Stephen Boyd’s lecture series where his printed slides are filmed from above, works much better than power point style lecturing.

      I dont understand why it is a problem to upload 6 x 10 minute sections to youtube. That will also provide an elegant meta navigation feature. I find youtube much more reliable than the streams on e.g. or Channel 9.

  8. Oliver Obst says:

    videolectures is great because it’s free to watch, and all is one place. But if cost is an issue, it may be an alternative to have to pay a small amount for watching the lectures, rather than having the authors pay more, or the attendees of the conferences. If there was a way to keep free access to attendees of the conference…
    With an average of 70 viewers, this makes approx. 3 Euros for an hour to cover the 200 something.

  9. Kamalika says:

    I agree with Oliver. I would definitely pay a monthly or yearly fee to subscribe to, or even pay per lecture.

  10. Jurgen says:

    “A videolecture is simply a new publishing format.”


  11. Although I never had the patience to listen to a full talk on videolectures, I agree that it is a value-adding feature.

    The cost is increased a lot mainly due to the need to have a whole team traveling to the conference (think, at least as many people as the conference parallel sessions + a few more for overhead tasks), and then have another team post processing the video, to make it available for easy online viewing. Once I saw the raw costs, I realized that the quoted “outrageous” price is actually pretty reasonable.

    Think: KDD will have approximately 2.5 days x 8 hours/day x 4 parallel sessions, ie. approximately 80-100 hours. Now the costs: To cover a 4-session conference, you need 6 people (4 people at the rooms + 2 people to run around and coordinate). So we need tickets + accommodation for 6 people just to cover the travel costs. Easily $8-10K, especially for transatlantic trips. Then add the salary of the 6 employees in the conference for the weekly trip. Another $2K, conservatively speaking. Then a few more days for post processing, another $2K at least, assuming that each hour is processed at a cost of $20/hr. This gives a low cost bound of of $120/hr. I would like to assume that they are not fully booked all day long, so we need to account for the “down time” of the business. Consultants assume that 2/3 of the working hours of the year will not be billable. Make it a a more conservative 50% utilization, this is a $240/recording hour. Here you are, with very conservative cost estimates.

    Now, having said that, the idea of a subscription to videolectures (e.g., an institutional license/subscription), seems to make a lot of sense. The ones that should be paying for the privilege are the direct beneficiaries.

    • “Easily $8-10K, especially for transatlantic trips.”

      This is one of the major points that irritated me when I first saw their price breakdown. It would be much more efficient to have at the very least one base in North America and one in Europe to more than halve the travel costs.

      • I do not think it is easy to maintain multiple offices, with employees sitting around in multiple continents.

      • Igor Carron says:


        I have seen higher fares from Houston to Paris then from SF to Boston. I can’t imagine the how the cost would be sensibly different with destination like Wisthler. in all, I would bet the total cost is likely to be insensitive to that part of the budget.

        The real cost are the man-hours, the delicate indexing of the video with the slide numbers and so on.


  12. Dave Lewis says:

    Agreed that this is simply a new publishing format, and so the “who pays” and “who benefits” issues arise. In contrast to conference papers (where costs are now so low that pure open access is arguably the only defensible approach), there’s some truly high costs to pay in 2010 to provide videoed conference talks.

    But we should keep in mind that in 20 years someone can plop down a little robot camera that will do the whole job (including editing). So we should avoid locking ourselves into a permanent paywall framework that we’ll need to fight our way back out of in 2030. For now, viewer pays seems reasonable, but there should be an exit plan.

  13. Lev Reyzin says:

    I think Oliver and Kamalika got it right.

  14. Tom says:

    Amusing video: “So you want to get a PhD in theoretical computer science?”

  15. I’m not sure you’d want to take care of all the filming, cameras, etc. etc. yourself, but in case you would, could be an alternative forum to publish the video data collected. An example:

  16. Ross Gayler says:

    I suspect you could cut the cost if you were willing to drop some of the components. videolectures looks as though it has been approached from a professional video production viewpoint, but do you really need everything it brings?

    Redwood Centre for Theoretical Neuroscience store all their seminar videos ( at the internet archive ( As far as I can see, upload and download are free. To the best of my knowledge the Redwood videos are shot on a standard consumer-grade camera operated by a graduate student or whoever is to hand at the time of the seminar. These videos don’t have the synchronised slide display of videolectures, but like Lars Pellarin I prefer to download the slides and manually browse them in parallel to the video.

    I find that the weakest technical aspect of these videos is the audio, especially questions from the audience. If you had a limited budget I would invest it in wireless microphones (lapel for the speaker and a roving hand-held for the audience) and having a pushy graduate for each session to insist that the microphones are set up and used properly.

    I also find that a video of the presentation is not any more useful than an audio-only recording synchronised to the slides. In fact, a poorly shot video with insufficient resolution of the slides is less value to me than an audio-only recording synchronised to the slides. The example of this type ( given by Arien P. de Vries seems really acceptable to me.

    I suspect there must be software that could run on the presentation computer to record the video out and audio from the line-in socket – so by patching the seminar sound system into the computer you could record the slides and audio presentation without needing a camera or camera operator.

  17. Nikos says:

    There is also another way. If a videolecture is a new publishing format, it is in the author’s interest to make a video of the talk. A conference can then have an optional video camera ready deadline, when the authors can upload a talk and their slides. Optional here means that the paper will be published regardless of whether the author uploads a video. A similar idea was proposed a couple of years ago by Vijay Vazirani on Lance’s blog.

  18. Eli says:

    YouTube has different partner programs..

    and for for educational institutions (scroll to the bottom and there is an “apply” link):

    They also have more general business partner programs. These all allow one to upload longer videos.

    Bandwidth costs are an issue if one decides to host videos themself.. (unless you are “lucky” enough to have no interest in your video lectures.)

  19. Zach says:

    Vimeo is free and has an upload limit of 500mb. Cloudera uses it for the videos from Hadoop World. Each of your speakers could also upload their slides to Slideshare?

  20. eugene says:

    I wonder if it is good idea to create videolectures-like website with videos/slides uploaded by users = some kind of user-generated

    This can reduce costs dramatically since video and slides will be prepared by users/authors/conference organaizing commitees.

  21. From the videolectures site – ‘The portal is aimed at promoting science, exchanging ideas and fostering knowledge sharing by providing high quality didactic contents not only to the scientific community but also to the general public.’ Since videolectures is ‘free’, at least to the public, there is a massive free rider issue. I had thought that most of the cost was borne by the EU, as a matter of public policy to provide a public good. I guess it is open to debate if there is a public benefit to providing this content, but as a non academic – and a free rider ;) – access to the content on videolectures is amazing and is an example of how the web can be of real societal value.

  22. Anonymous says:

    VideoLectures is a nice idea but the people running it should fix the bug in the algorithm measuring authors’ popularity; otherwise, not only me, but other people will become suspicious of what’s going on…

  23. The fact that NIPS 24 is now two weeks old and there are no lectures up at alone suggests that it is not worth it. Most meetings we run with video in astronomy (there aren’t many) have video and slides up same-day. Of course if were a non-profit archive running in the public interest with a long-term commitment, I might change my mind. One of the university libraries could serve this role, as Cornell library does for

  24. Anonymous says: will a good alternative to

  25. Swaprava says:

    Though I’m not sure about what policy the conferences and the institutes follow for their talks and courses, it is certainly becoming very popular and useful to improve the quality of knowledge and research in the entire community. For example, I listened to Andrew Ng’s Machine Learning Course of Stanford and Satinder Singh’s Tutorial on Reinforcement Learning and both of them were very useful. I can certainly say this is becoming very popular to disseminate, sometimes clarify, ideas. It is of course a matter of debate why it should be made available publicly to people who didn’t pay for it. But, keeping the monetary part away from the discussion, it synchronizes the thoughts and even jargons and notation for the community working in the same area. As a huge fan (and contributor sometimes) of opensource software, I strongly feel this dissemination of knowledge would certainly help the students over the time. I would like the experts to settle the registration/ non-registration issues and the payments but want them to keep the benefit of the community through dissemination of information in mind.

  26. Piotr says:

    Speaking from a student perspective (so another kind of a ‘free-rider’), I can say the free-access model is a very neat thing of It often gives access to quality stuff on topics not yet covered in handbooks and in more accessible format then research papers. My first contact with some of the ideas on modelling in life sciences was on After that I reached for proper publications, of course.
    Not all stuff submitted to various video-hosting web pages are in watchable quality, so this $270/hour spend on professional crew are usually well spend money. And it’s still a bit pioneering kind of publication format (not really popular in less computer-dependent fields of research), so decision-making people yet need to be convinced, it’s worth their attention. Until it won’t reach the mainstream (e.g. video lecture cameras will become parts of lecture halls multimedia projectors), it will be a bit pricey, I’m afraid. I see Dave Lewis perspective – maybe it’s worth going through this bumpy period for the sake of future benefits?

Leave a Reply

Powered by WordPress