Last semester saw a fairly successful run of TCS+, the online seminar series that Anindya De, Oded Regev and myself co-organized over Google+ hangouts. We’re very pleased that all nine seminars were very well attended. The seven “inside spots” almost always filled up; our most loyal supporters included groups from University of Michigan, ETH Zurich, NYU, Columbia, Berkeley, TTI & University of Chicago. In addition we typically had around 30 viewers connected to the live stream at any point throughout the talk (though there’s no telling whether this number was generated by a constant turn-around, or if people tended to stick around for the whole talk — hopefully the latter!). The TCS+ YouTube channel registered 3,222 page views and 28,269 minutes of video viewed (including 1129 minutes, or about 19 hours, during the month of August, which I’ll interpret as proof that archiving videos of past talks *is* a valuable service — there are an increasing number of such archival services available on the web; see here for a partial list). All in all that’s about 471 hour-long talks, a healthy amount of attention shared between our nine excellent speakers, whom I wish to thank once more for their patience and effort in making this work: Ronald de Wolf (whose inaugurating talk, pictured above, remains the most viewed on our YouTube channel), Anup Rao, Raghu Meka, Nisheeth Vishnoi, Greg Valiant, Alexander Sherstov, C. Seshadri, Thomas Rothvoss, and Jelani Nelson. Feedback on the TCS+ experiment has so far been overwhelmingly positive, and I thank all our viewers for their support!
You’ll have noticed that both our past choice of speakers, and the groups that have been joining us live, are quite US-centric. This is definitely not deliberate, but rather an unfortunate consequence of the way we organized the talks: all three of us were at US institutions, meaning we’d more easily have contacts with US-based speakers; in addition scheduling the talks at a time when people on the West coast would (more or less) be awake meant it would already be early evening in Western Europe, and nighttime in Asia. I view this as unfortunate because the issue of distance between speaker and audience is is exactly what TCS+ is meant to overcome. We’d like to have talks by US speakers that are widely viewed by researchers at European or Asian institutions, and vice-versa. There is clearly demand for this: guess which country has been contributing, by far, the second largest number of viewers of TCS+ seminars: India! They are a fairly anonymous crowd, watching talks scheduled at 10 or 11pm from the comfort of their dorm room. But they are precisely the viewers we’d like to be reaching out to. So we’ll hopefully try for a more balanced schedule this semester. This will in part be achieved by shifting the organizing team’s center of mass to the East. As a first step towards this, I transferred my own moderate mass from “new” to “old” Cambridge. More importantly, we are very pleased to be joined by Thomas Holenstein from ETH Zurich in the organization of TCS+. This should let us organize some talks at a more reasonable time for our Asian viewers, and we of course also welcome suggestions of speakers from European or Asian institutions!
The TCS+ seminars have not been without a few hiccups, including a scary fire alarm at MIT, forcing me to abandon my laptop (which was hosting Ronald de Wolf’s talk) in an empty room at MIT while we were evacuated (you can still see the flashing lights in the online recording), and a network outage at MSR New England just 10 minutes before Greg Valiant’s talk, prompting him to dash for the nearby Stata center in order to recover connectivity (luckily the outage was short-lived and Greg could deliver his talk safely from MSR). All in all though, the “hangout” feature of Google+ has served us well, proving both remarkably stable (0 crashes) and easily accessible. Some groups did sometimes have trouble joining, but except maybe in one or two occasions everyone who tried hard enough eventually made it relatively unscathed.
There remain two main unsatisfactory aspects of the experience. First, the video quality is not great. The hangouts force a particular layout in which video feeds from online viewers take up a good chunk of the screen. In addition, it often proved difficult to setup the presenter’s computer so as to efficiently juxtapose his slides and a video feed of himself (and I certainly wish I could add “or herself” here: we’re definitely behind, not only in geographical balance, but also in gender balance…let’s work on this!). Second, the hard limit of 10 online viewers, though not absolutely crippling thanks to the online feed, does result in “outside” viewers feeling not as included in the talk as they should. This maybe partially explains why we did not have as many questions, and interaction, as we were hoping for. (Any suggestions to improve this?) While the “online” groups regularly asked questions, very few “outside” viewers used the comments feature of G+ to ask questions. Indeed, the delay in typing up the question, together with the necessity for a moderator to spot the comment, read it, (optional) understand it, and interrupt the speaker to ask it, made it a fairly useless feature.
So we explored around a little bit for options. The best alternative so far, suggested by Thomas Holenstein, has been a videoconferencing software from Radvision called Scopia. Scopia allows any number of simultaneous participants. It also makes it easy to show both a video feed from the presenter and his slides. Unfortunately so far we are not fully satisfied with the video, and even more sound, quality that we’ve experienced; this seems to be especially so when some of the participants have low bandwidth. Trying out alternatives made us appreciate how good the sound quality is in the hangouts, with very little interference or losses. Moreover, Google just announced that live streaming would now be available in HD, meaning that the quality for our outside viewers, and for the recorded video, will be much higher. The only remaining limitation is the 10-slot maximum. There seems to be a possibility to extend this to 15 slots, though we’re having trouble getting in touch with the right person at Google to help us obtain this extension (which we’d be happy to pay for). Anyone can help?
Another slight negative of using Google+ is that, well, not everyone uses it. This means that some of our talk announcements might go unnoticed; we’ve been trying to forward them to university mailing lists as well but it’s hard to keep track of everything. Though we’ll continue to use the TCS+ page as our main forum for the foreseeable future, we also created a more traditional mailing list to which we’ll send out announcements. This will be a low-volume list, with maybe 2 messages per week (an early announcement, and a reminder, for every talk). You can subscribe to the list now by clicking here. If there is any way announcements from the list can be forwarded automatically to the appropriate list at your institution, please let us know how! Our website also includes a Google calendar and RSS feeds that you can subscribe to.
I’ll conclude by announcing the first talk of this new season: we’ll have the pleasure of hosting Ramprasad Saptharishi, from MSR India. (Ramprasad agreed to talk at our usual 1pm EST, which will be 10:30pm for him, meaning we already wasted our first opportunity for shifting our schedule East; let’s do better next time!). Ramprasad’s talk is titled “Arithmetic Circuits: Depth reductions, chasms and escalators”. He will tell us about a very strong threshold phenomenon about circuit lower bounds that him and co-authors have recently identified: while we do have number of reasonable lower bounds for circuits of depth 2, strong enough lower bounds for circuits of depth just 3 would already imply lower bounds on general arithmetic circuits, a holy grail of complexity theory. Please join us for this exciting seminar!