Well, it feels that way at times. NetSci 2006 has been a great experience, exposing my designerly brain to new ideas in a relatively new field of network science. I missed my ride(s) in on Friday and opted out on Saturday, giving me several days to rest up for more information overload. So far, the sessions have distributed pretty consistently into 1/3 great, 1/3 good and 1/3 my-head-is-exploding. Monday — the first day of the actual conference, with the workshop officially ending on Saturday — was no exception.
There were a lot of topics covered (14 speakers, plus almost two dozen posters). Here were the highlights:
- Epidemics / Alex Vespignani (Indiana) — Alex reiterated some stuff from the workshop last week, but new (to me) was the range of modeling strategies from Simple to Realistic. This spectrum starts with Homogeneous Mixing (everyone is connected to everyone else), to two conceptual models (Social Structures and Contact Networks), up to two data-driven models (multi-scale and agent-based modeling). Since I’m still building my vocabulary, that was helpful information.
- Network Kriging / Eric Kozacyk (Boston) — I had to look up kriging on Wikipedia. Didn’t help much, but the central topic of Eric’s talk was about monitoring path traffic in a network using a significantly smaller sampling. Although very mathy, I did get the basic idea that precise monitoring is too costly, and this kriging method leverages the redundancy of major paths and ignores the single-degree nodes that attach to those paths.
- CiteRank / Dylan Walker (Stony Brook) — Google page rank methods were applied to a network of academic paper citations for physicists. This was an attempt to overcome biases in the current method of citation rankings, the number of citations received, which favors older papers and ignores recent importance and popularity. Other than having an existing network to play with, I’m not sure what the practical value is for this. Is it something that will help researchers find relevant stuff, or is it a value mainly for well-positioned authors trying to get tenure or raises? Regardless, interesting talk.
- Price of Anarchy / Hawoong Jeong (KAIST/Korea) — Although the study itself seemed a bit green, this work was setting up the question of how we can construct of a better network that decreases Price of Anarchy. PoA is the cost a decentralized network pays for not being coordinated. The example of this concept considered two roads between destinations, a toll road and free back roads. The problem is that the initial optimization is thwarted by the toll-road travelers continuing to migrate from the clearer highway to the congested back roads to save personal cost … and in the process, increase collective cost. Hawoong examined the Boston Road Network with 59 exchanges and 108 road segments. The conclusion suggested the city would be better off eliminating some roads.
- Self-Similarity in Growth of Complex Networks / Hernan Maske (City College) — A bit mathy, but this talk looked at a way to measure fractality by drawing boxes around nodes within a given distance (much the same way grids are used to estimate the fractality of coastlines). The big idea here was that this is a way to “predict the past” by renormalizing a network and discovering its ancestral node. Nah, I don’t understand it, either, but I can’t help but think it might be relevant at some point in the future.
- IPD Network Creation Games / Jan Scholz (Goethe/Germany) — This guy looks an actor I can’t place, one of the junior bad guys from some T.V. show I can’t remember. He’s also the speaker who has come closest to a really well-done power point slide set. Easy to follow and not too much on the screen, something that plagues most of the other efforts thus far. IPD is “iterated prisoner’s dilemma,” a game theory idea in which agents encounter other agents and try to improve their own reward through linking strategies. The “Nash Equilibrium” turns out to be where all players defect. In the iterated version, however, a sense of memory is added where agents remember what strategies others have taken in the past. This change allows eight strategies to emerge. The point is to avoide expensive optimation by allowing nodes to play games with each other.
- Modularity / Mark Newman (Michigan) — Another nice talk from Mark, the focus this time was on using the measure of modularity to understand more about the community structure of a network. Communities are identified when you can look (or calculate) at a network and figure out how to cut some edges to show the separation. Sometimes it is easy; sometimes it is not. A high modularity value indicates a good division. Or the inverse can be examined (the “negative eigenvector”) to try and create a bipartite network, where there are no edges within a community. In this case, the minimum modularity is the indicator of good division. I also picked up a blog to investigate after asking Mark about a slide he showed on political books.
- Large Maps of Science / Kevin Boyack (Sandia National Lab) — This is another one of those “So, what is this good for?” topics: mapping science. Kevin gave a strong talk explaining the history of such maps, from 1974 to 1999, before sharing Sandia’s map. There’s a lot I don’t understand here, but I was intrigued by the notion that this might be used to generate topic maps from forums or even create some kind of meaningful summary of content. I think that is an important need for most large forums to increase a new user’s accessibility to the cumulated history and knowledge shared.
There were also a number of interesting posters, many from Indiana. My favorites included a thin preliminary study that examined basketball passing as a network; GiveALink.org, which uses dontated bookmarks to build search relvancy; a study of United Way funds that looks to find gaps in the use of dontations and their effectiveness in getting to the target population; visualization of music libraries that unmask small listener communities; and age-related differences in MySpace networks.
Tuesday may prove to be a day weaving active with passive work, as I still need to fit in some time to finish some web sites and wrap up capstone stuff. Lots of math and bio on the docket, so it’s probably a good day to mix it up.
