Much Belated Entry

[tiedyedave]

Belated is hardly the word; I haven't LJd in over four months.

Hi there!

So stuff happened during the past four months. This stuff,
with some loss of generality, can be described thusly:

I took classes. I did pretty well. Taking Artificial Intelligence and Cognitive Modeling at the same time was a very educational study in contrasts. I fared poorly on finals because of my weakness for Diablo II, but it turned out all right.

I got a summer job. Go research! For any of you who wonder what research is like, it's all about the papers. If you can read three papers a day on ten different fields you have no background in, and understand them, you're a good researcher. Results will follow naturally.

I can't do any of the above because I'm l4m3. But it's fun to try anyway. Multicast networks are fun. :-)

I've been reading my friends' entries. You know what? A lot of my LJ friends are people I don't know too well irl, so I never really think much about what their lives are like. But I've been reading your entries lately, and you're all cool. Profound. Human. I believe that LJ is a great way for people to communicate themselves to each other because it gives you the chance to reveal yourself thoughtfully and gradually in ways that everyday life sometimes doesn't allow.

Just sayin'.

Comments

[fleaplus.livejournal.com]

So what sort of research are you doing?

[tiedyedave]

I'm actually in a couple of different things.

My main project is mechanism design for multicast networks. Suppose you want to multicast a movie over the Net to a bunch of users. It costs you money to cross links, and the users pay you back for the service.

Consider a graph G, where all of the edges and vertices are 'players', except for one 'root' vertex, which is you. Each edge and vertex wants to maximize the amount of value it gets by playing the game, and you want to build a tree from the root that gets you as much money as possible (or at least lets you break even). So you have to design an algorithm that builds this tree, but you don't know how much the edges cost and how much the vertices value the service; they can say whatever they want. Thus, the algorithm has to force truth-telling to be the best strategy for everyone.

Vertices as agents with known edge costs has been solved.
Edges as agents with zero vertex value has been solved.
Right now I'm working on edges as agents with fixed vertex values.

I'm sorry if the description is not very clear, but it took me days of reading papers and talking to Amitabh just to get a handle on the stuff. :-)


The other project is an image classification game; Ian and I are trying to write a little game that gathers information about images by using human players as 'oracles' and still making it fun.
The fun is the hard part. :-)