Log in

No account? Create an account
Talk videos - Arvind Narayanan's journal [entries|archive|friends|userinfo]

Talk videos [Apr. 22nd, 2007|02:07 am]
Arvind Narayanan
[Tags|, ]
[Current Mood |contemplativecontemplative]

Vijay Vazirani's recent post about video submissions made me think about the obvious, noncontroversial things that we're not doing. I suggested we should start recording conference talks and put them on the web first, to which I got the reply that this was a "no-brainer." Well, then, it's all the more hard to understand why it isn't being done.

The majority of technical paper presentations are not at conferences. Rather, they're given in small rooms of perhaps a dozen people at universities/corporations. However, a paper presented locally is by no means a worse presentation than one at a conference (in fact, it's often a lot better because you can't cover much in 20 minutes at a conference.) Why, then, aren't these talks published?

One answer is that corporations probably view talks as IP and therefore won't share them with the world; that's understandable. And universities can be behemoths when it comes to these things; setting up video equipped rooms and scheduling talks can take forever to get past the the red tape. So, why not take matters into our own hands?

The incentive is there: a paper with an accompanying video is sure to get much greater visibility. The cost is pocket change: everyone and their dog has a webcam these days. The quality is going to be shitty, but what do you really need more than webcam quality for?

So here are my questions:
  • Are there any CS departments where talks are routinely put on the web through official channels? (Kind of like MIT opencourseware for research talks.)
  • Is there anyone doing it on their own?
  • Do anyone know of publicly available software to integrate the video with the slideshow? I've listened to a talk in this manner and it's super-awesome: you can navigate based on either the slideshow or video and it stays in sync automatically.
Coincidentally, some commercial efforts targeting "cerebral" uses of online video have been in the news recently. (For instance, fora.tv.) These aren't crucial, since we all have access to good hosting bandwidth, but it's nice to know they exist; it probably suggests that online tech talk video is about to gain critical mass.

[User Picture]From: brokentooth
2007-04-22 09:36 pm (UTC)
> Do anyone know of publicly available software to integrate the video with the slideshow?

One of the PhD students in the vision lab here is working on matching slides to presentation videos automatically. He's developed some really cool software which isn't publicly available yet, but should be, soon.

MSRI has a lot of lectures online: http://www.msri.org/communications/vmath/index_html. They had a series of vision workshops in 2005, which I found to be *extremely* instructive.
(Reply) (Thread)
[User Picture]From: arvindn
2007-04-23 02:48 am (UTC)
Well, the slideshow-video stuff is already deployed internally at Microsoft for all their talks. So the technology exists.. although I don't know if their thing is automatic or requires a human to specify the mapping. You can't tell the difference as a viewer.

Thanks for the MSRI link.

(Reply) (Parent) (Thread)
[User Picture]From: ephermata
2007-04-23 12:14 am (UTC)
The University of Washington actually televises many of their lectures on "UWTV." Imagine my surprise when I turn on the TV a few weeks ago and see a talk about PlanetLab! There are a bunch of talks online:

As for doing it ourselves, that's a reasonable idea. Probably the quickest way to get started is for the speaker him or herself to bring the camera and ask a volunteer to do it. Then upload to YouTube. Yes, we all have access to good hosting bandwidth, but this is about doing things with a minimum of setup so more people will participate.
(Reply) (Thread)
[User Picture]From: arvindn
2007-04-23 03:04 am (UTC)
Oh, I have nothing against youtube. Its just that I thought it would make the material harder to find, because the overwhelming majority of youtube content is non-academic. But that's a minor point. The important thing is to get the stuff out there.

UWTV is very interesting. I missed Tapan Parikh's talk at UT, and it's supposed to be internally available at UT; I tried to find it but failed miserably -- didn't even know where to start since I couldn't google it. Now I click on your link and Parikh's talk at UW is the first talk on the front page! Problem solved :)

I guess there are two lessons here: internal distribution seems to be useless, because even internal people can't find it; secondly, in the last couple of days I've found 4 or 5 excellent sources of research talk videos that I never suspected existed. I really think we need some kind of indexing system. Something that can be used as an oracle for the question "given a paper X, does there exist a publicly available video presentation of X?"
(Reply) (Parent) (Thread)
[User Picture]From: ephermata
2007-04-24 04:38 am (UTC)
Well, Youtube is more easily searchable than a collection of scattered internal video sites, which may or may not be open to the external world. The idea of having a search just for talk videos is not bad. The quick and dirty thing to do would be to start with sending a search to YouTube for keywords and reporting the results...
(Reply) (Parent) (Thread)
[User Picture]From: arvindn
2007-04-24 05:23 am (UTC)
BTW, youtube has a 10 minute cap on video length. They have a keyword search already, but the problem I'm trying to address is indexing the content that's already out there on various sites, and will probably never move to youtube. UWTV alone is enormous. I'm impressed, amazed. There's way too much content to wade through manually.

I'd say a quick-and-dirty solution would be to scrape all these pages and upload their URLs to del.icio.us with a common tag like say "techvideo" and appropriate subject tags for each URL. What do you think?
(Reply) (Parent) (Thread)