EC 2010 (nominally stands for “Electronic Commerce”) was recently hosted by a dedicated team at Harvard. I had a great time as an undergrad there and it was nice to be back for a few days, revisiting places like Felipe’s and People’s Republik. The conference itself took place in the skylighted underground piazza of the Northwest Science Labs, which was really the ideal space for a small conference. The building itself is an enduring symbol of the Summers era at Harvard — shiny, science-focused, and intrinsically endowed with the hubris that comes from constructing such a costly structure without even bothering to secure a naming donation.
My favorite talk was from Ramesh Johari for “Congestible services and network effects” (joint work with Sunil Kumar). Unfortunately, the paper does not seem to be available openly online and even the ACM Digital Library copy is just a one-page “extended abstract” (UPDATE: the paper is available here). Despite the awfully throwaway title, the work is a clever analysis of two opposing factors in networks: network effects, people benefiting from the presence of others (think Facebook), versus congestion effects, where your utility decreases as more people use the service (think traffic). I disliked the use of Nash equilibrium in the paper, but from what I could tell it looked like these were potential games and so simple dynamics should reinforce these equilibria. A robust model drawn in broad strokes and with qualitative insights — to me, the ideal theory paper.
The other talk I particularly enjoyed was from Sharad Goel for his paper (with Dan Reeves, Duncan Watts, and Dave Pennock) “Prediction Without Markets”, an analysis of just how well prediction markets do their job by comparing against alternatives. I just really liked how much data went into this project — the authors evaluated so many distinct datasets and went into a good deal of depth to contrast various predictive techniques. I thought the prediction vs. probability graphs with circles to reflect the number of samples were a clever way to present a complicated data analysis cleanly.
