I’m at ICALP 2010 in sunny Bordeaux. I have been very busy working on two papers with colleagues who happen to attend ICALP as well, so I have missed a bunch of interesting talks on the first two days already.

I did, however, attend the business meeting, an event that combines various reports from ICALP committees with the general assembly of EATCS, the “European SIGACT.”

I normally enjoy business meetings, but ICALP business meetings are relatively stiff and mind-numbingly boring, with no discussion and no free beers. Still, let me give an incomplete and skewed summary of some of the issues that were mentioned.

ICALP publication

EATCS views ICALP as its flagship conference, and ICALP has published with Springer’s LNCS proceedings series for many years – if DBLP is to be trusted, the proceedings from the 4th ICALP (Turku, Finland, 1977) appear as LNCS volume 52. Since a few years ago, Springer has started an LNCS subline called ARCoSS (Springer page about ARCoSS) “in cooperation with” EATCS and ETAPS. These have a nicer cover.

However, it has been unclear in which sense this publication model coincides with the current preferences of the members of EATCS. Just to mention an alternative, the STACS conference now publishes with the LIPIcs series, which is Open Access in the sense that they are available on line and free of charge.

Very commendably, and to my pleasant surprise, EATCS actually implemented a member poll in the Spring of 2010 to determine our opinions on open source publication models, and in particular the outlet for ICALP’s proceedings. The results are online ([poll results at EACTS website]), but for mysterious reasons only accessible to EATCS members. It’s fair to say that there is a large majority of members that would prefer another publication model. For example, only 10% want EATCS to “Publish the proceedings of ICALP as before in a traditional print publication?” To the question “What is the publication model for scientific papers in theoretical computer science that you would prefer to see gain prominence in the near future?”, 83% respond “Open access publication as for LMCS, LIPIcs and EPTCS”

EATCS chairman Burkhard Monien reported that the EATCS Council has now approached Springer about the ICALP proceedings. According to him, Springer has made substantially better offer “very close to open access. We have discussed this carefully in the council.”

I think this is good news. However, I am less than impressed about the speed with which the process will continue. According to Monien, the result of the May 2010 poll were not sufficiently conclusive, so “We’ll ask your opinion again.” This time, there will be mercilessly concrete choices about either (1) going with LIPIcs, or (2) staying with Springer, under specific (and, I assume, new) terms. Monien explained the EATCS council wants to get this right, ICALP is a big steamer, should not go in zig-zag course, and stick to its decisions. He was unsure if the next poll will be implemented in the Fall, but promised “beginning of next year, at the latest.” I’m not holding my breath.

Still, all in all I think this is splendid news, and I’m particularly happy to the the EATCS council involving its members in an important decision process.

ICALP 2010 Organization

The report from the ICALP 2010 organizers was the usual list of statistics. I have become increasingly interested in these details after arranging ALGO 2009 myself. In total, ICALP 2010 has 292 participants, including 55 locals (the latter strikes me as a high number). 200 of these are registrants for ICALP itself, rather than one of the affiliated workshops, 147 pay the full registration fee. Only 3 Swedes and 1 Dane are registered, I wonder which I am.

This ICALP is held in the conference facilities of a large, soul-less hotel in down-town Bordeaux. I’m not a big fan of this kind of conference, and the contrast in atmosphere to the locally organised SWAT I attended in June is enormous. There’s a small lobby to hang out in, with few chairs or tables (let alone blackboards), interaction between attendees is minimal, except for the cliques one is already in. Furthermore, ICALP provides no lunches, so people scatter into small groups and look for a place to eat. Pretty much like the large US conferences.

Also, the Friday banquet is included only for full participants, meaning that many students will not be able to get the 90 Euro dinner ticket reimbursed, and consequently won’t attend. In my experience, this decision is de rigeur for Southern European ICALPs, but I think it is fundamentally misguided. The food better be spectacular.

On the other hand, the proceedings are included in the registration, which is another mistake. Many people never even remove the shrink wrap anymore, paper proceedings are just ballast. Moreover, proceedings are easy to reimburse, even for those students who honestly prefer to pay for the privilege to read the absurdly formatted 12 page LNCS amputated paper rather than reading the full version on the web.

The budget for the whole event is slightly over 100,000€, and the organizer received 2,500 emails. Apparently, EATCS maintains some for of organizer’s manual for ICALP; I’d like to see that. (I have seen the ESA/ALGO manual.) These things should be on-line, and allow for feed-back.

Programme Committees

The reports from ICALPs various programme committees included the usual absurd statistics, which were useless even by the low standards we are used to. The best paper award (Track A) went to Leslie Goldberg and Mark Jerrum’s result about about Approximating the partition function of the ferromagnetic Potts model. This is, of course, utterly splendid work. In total ICALP Track A received 229 submissions (6 were withdrawn) and accepted 60. Denmark submitted 3.33 papers, 1.17 accepted. Sweden submitted 1.83 and accepted 1.17. Again, I wonder which I am.

The programming for ICALP’s Track A has left me, and many others, completely puzzled, and looks as if it’s the result of a grep script based on paper titles. For example, ICALP had two strongly related results about Cuckoo hashing; but the results appeared in different sessions:

• Martin Dietzfelbinger, Andreas Goerdt, Michael Mitzenmacher, Andrea Montanari, Rasmus Pagh and Michael Rink’s Tight Thresholds for Cuckoo Hashing via XORSAT appeared in session 3 on “Sorting & hashing,” inconceivably scheduled in parallel with the session 3 on “Data structures”. These are the only sessions on data structures in a 5-day, 3-track conference! In parallel!
• Nikolaos Fountoulakis and Konstantinos Panagiotou’s Orientability of Random Hypergaphs and the Power of Multiple Choices appeared in session 5, “Graphs and hypergraphs.”

ICALP 2011

Michael Hoffmann presented the ICALP 2011 organisation in Zürich. icalp11.inf.ethz.ch . This seems to be an ICALP model closer to my preferences. The conference will be held in in university facilities, at something called CAB from late 19th century, with several large lecture rooms for workshops or parallel tracks, an even larger lecture hall, WLAN, and plenty of working space.

Also, there is no banquet, but rather a number of receptions with plenty of finger food and drinks. As an added benefit, registration fees are expected to be low. Of course, Zürich is “not among the cheapest places”, and prices are even higher than shown last year, because of the exchange rate between Swiss Franc and Euro, which shows high variance.

I will certainly attend, as I am one of the invited speakers. Apparently, I’m speaking on the 5th.

ICALP 2012

Artur Czumaj presented the plans for ICALP 2012, which (somewhat unusually) is already fixed two years in advance. The dates are July 9–13 2012 in Warwick, for the centenary of Alan Turing’s birth. Thus, ICALP is one of many events in the UK for the Alan Turing year, see the list of events. ICALP will be held on Warwick’s campus, including accommodation and lunches. Satellite events were said to include Wimbledon right before ICALP and the London Olympic games somewhat later.

Possibly this is the right time to start thinking about what we in the greater Copenhagen area (e.g., Sweden and Denmark, or all of Scandinavia) should be doing for the Turing year. Computer Science has far from the pop-sci clout that biology enjoys, but shouldn’t we be able to drum up at least one hundredth of the attention that Darwin got? Ideas are welcome.

For example SWAT 2012 is in the Turing year. We should start branding all our CS-related activities in 2012 as Turing-related, and hopefully concoct some more.

The SWAT conference dinner involved a boat trip to an island near Bergen, with a spectacular seafood dinner of Halibut and Wolffish. Also, initiated by Fedor Fomin, SWAT reverted back to its tradition of pressuring attendants to sing (partitioned into small groups by nationality or country of residence) after the main course. Again, the organisers managed to get the atmosphere exactly right.

Invited talks

1. On Monday, Sanjeev Arora talked about semidefinite programming and approximation algorithms. A lot has happened in this field since what he called “first generation” SDP algorithms. Some of the most recent works concerns re-evaluatating the role played by the Unique Games Conjecture.
2. On Tuesday, Prabhakar Raghavan, now at Yahoo! Research, talked about quantitative models for user behaviour. He spent some time on models for presenting search results (in particular, images) in two dimensions – which is much less obvious than the top-to-bottom ordering that web pages are presented in. (Left-to-right, row-by-row, is not the right answer for laying out pictures in a grid, because the eye doesn’t move like that.) He used this example to also communicate broader points about the interplay between quantitative sociology, cognitive psychology, and theoretical computer science. Great stuff, and highly interesting to me both because of the “algorithmic lens” propaganda, and because I sometimes give general audience talks about the computer science behind search engines, social networks, etc.
3. On Wednesday, Dana Randall talked about phase transitions in sampling algorithms and underlying random structures. This was right in the middle of the interplay between statistical physics and theoretical computer science that I am currently fascinated by. The talk was an algorithms-friendly introduction filled with rapidly mixing Markov chains, crisp combinatorics, Ising and Potts models, and colloids (which I hadn’t seen before.)

I am attending SWAT in surprisingly sunny Bergen.

The biannual SWAT is the oldest European conference devoted to algorithms. (ICALP is a lot older, and STACS slighly, but both have a wider focus.) It rotates around the Scandinavian countries (with a single exception in 2006) and in 2010 Bergen is, again, the host.

Bergen has a large algorithms group with a bunch of postdocs and graduate students, and this makes a large difference for organising a successful conference. A lot of heart and energy has gone into this, and the atmosphere is absolutely splendid. Even the weather in Bergen seems to, so far, play along.

The Sunday evening reception was on a roof terrace, with live music, wine, and snacks. Several activities were arranged for Monday evening, and I was lucky to be among the handful of people who went sailing on a small sailing boat, late in the evening, under the Northern sun, with a view of beautiful Bergen and the surrounding coastline. Brilliant.

Business meeting

At the business meeting, Magnus Halldorson reported on the steering committee’s thoughts about SWAT. Four changes have been implemented at SWAT 2010:

1. Name change. SWAT’s name is now “Scandinavian Symposium and Workshops on Algorithm Theory,” rather than “Scandinavian Workshop on Algorithms Theory.” The “workshop” monicker in the original name has been a recurring problem for SWAT attendants who tried to secure travel money, looks bad on CVs outside of our narrow community, and even made it difficult for previous local organisers to attract outside funding.
2. Economy. SWAT registration fees are seldom cheap because of the high prices in Scandinavia, and Norway is currently extra expensive due to currency exchange rates. Still, the steering committee wants SWAT to remain economically viable for a large number of attendants. The local organisers managed to secure significant outside funding, so that registration fees only make up 30% of the total revenue. Most of the money came from the Norwegian Research Council and Bergen University.
3. Time slot. SWAT 2010 takes place around Midsummer, rather than the traditional week “before ICALP,” typically the first week of July. I think the SWAT 2010 date makes a lot of sense, but it is unclear whether this will be a long term change. A show of hands from the attendants at the business meeting was slightly in favour of the earlier time slot.
4. International conference. The steering committee made it clear that SWAT is an international conference with an international PC and no special attention paid to Scandinavian papers. (I note that this SWAT has a total of zero papers from Sweden and Finland.) Moreover, the programme committee and the organising committee are completely de-coupled.

I think these decisions represent a very welcome re-alignment of how SWAT views itself. The attendees of the business meeting seemed to be happy as well.

PC chair Haim Kaplan gave a short overview of the work of the programme committee. The most interesting point is that SWAT 2010 attracted 78 submissions this year, which is markedly less than for previous SWATs. This year, the SWAT submission deadline was before the notification of SOCG 2010 because of the June time slot, which may account for a decrease of the number of submissions in computational geometry, traditionally a thematic focus of SWAT.

A brief discussion was devoted to publication models for contributed SWAT papers; currently the conference is published in Springer’s LNCS series.

Petteri Kaski successfully invited SWAT 2012 to Helsinki.

This is part 3 in a series of 3 posts on ALGO 2009, largely repeating the organiser’s report from the business meeting. This is the least serious part.

Name tags

One of the small details that I tried to micro-manage are name tags. For some reason, name tags at many conferences display the attendee’s name in very small letters, and leave most of the tag’s area either white, or taken up by the logo of the conference or venue. (Professional name tag designers typically optimise something else than ease of identification of the attendee. That’s their job, and they are good at it. But it’s not the job I want done.) ALGO had nice, legible name tags, and we also tried to double-check various diacritics.

One the other hand, the batch-driven production from a spreadsheet to the final tag messed up some affiliations, for reasons that remained a mystery. Incidentally, I don’t really understand the value of displaying the affiliation on the tag anyway, they seem to serve mostly as an ice-breaker for conversations. (Maybe we should put people’s hobbies on the tags instead.) From that perspective, a wrong affiliation works at least as well.

Closely related to the layout of the name tag are the mechanics of the badge holder. After obsessing over this for many years I have concluded that the pin–clip combo is the way to go, so that’s what we got. We even told people to not clip the badge to their trousers. It was perfect, nothing could go wrong, everything was prepared for the one conference where you wouldn’t have to squint or use x-ray vision to read the name of whats-his-name-I-think-we-met-at-ICALP-last-year.

Alas, it was not enough. A seemingly unrelated decision destroyed all my careful planning: We decided to hand out multi-ride ticket to the Copenhagen subway system, attached to ITU-branded key bands. (This was to prevent the tickets from being accidentally lost in the pile of written material we handed out to attendees.)

The effect?

Driven by their desire to remain anonymous, and a perverse predilection for craftsmanship, many enterprising attendees carefully removed the ticket from the key band, and attached their name tag instead, thereby constructing the mothership of all that is evil about name badges: they dangled at belly button height, typically obscured behind tables or under jackets, and they swivelled around their single, ill-constructed joint, displaying the badge’s empty back side more often than the laws of probability theory predict.

Next time I’ll tatoo the name on people’s forehead.

Computer use

The etiquette about open laptops during talks is very much unclear in the CS community. We politely asked people to avoid accessing the internet during sessions, and if so, to do it from the back row so as to annoy the speaker and the rest of the audience as little as possible. Several attendees told me they were happy about this decision; the auditoriums were certainly relatively free of open laptops. I remain uncertain about how to handle this – I can get a lot of work done while half-way listening to a talk, and the alternative is to stay out of the lecture room…

Speaker instructions

One thing I decided against, for fear of committing a social mistake, was to send instructions to the speakers about how we would have liked the talks at ALGO. Namely, more accessible.

Talk quality at our conferences is generally high: clear, readable, well-presented, entertaining. However, most of the speakers vastly overestimate the audience’s level of familiarity with the subject. This makes it very difficult to attend a presentation to learn something, in particular about a field in which you aren’t an expert already.

One of the problems is that talks are grouped into thematically related sessions. This seems to imply that everybody in the room is already an expert about “latest derandomisation tricks in quantum I/O kernelisation,” so why bother explaining the basics again? An earlier rumination of Michael Mitzenmacher about the STOC 2009 schedule already touched on the bold idea of mixing sessions at random. (But this may introduce more problems than it solves.)

The smaller events at ALGO, such as WAOA and IWPEC, are technically highly specialised workshops in the first place, pitched at an expert audience. So how do we turn this around? It would be great if ALGO attendees could “stay for IWPEC” to finally find out what “this FTP-stuff” is all about – but the atmosphere has the exact opposite effect. Similarly, I talked to a WAOA attendee who was frustrated about the O-part being largely incomprehensible. (The attendee was there for the A-part.) How do we solve this? Clearly, not every IWPEC talk can define treewidth again, but my ambition with ALGO was to reduce balkanisation, not contribute to it.

Many of the “too difficult” talks are given by students. There are a number of reasons for this – students assume that everybody knows what they do (I did), students need to demonstrate that they’re smart and did a difficult thing, some students lack the broad overview to be able to place their subfield into a broader ensemble, previous talks have been given only at departmental seminars, where everybody else is equally interested in derandomising quantum I/O kernelisation, etc. Maybe for this target audience, a “Here’s what we’d like your talk to be”-letter could have done a lot of good.