Using Fundamental Economic Solutions to Solve Real-World Problems

点击翻译按钮获取中文摘要

The alliance between computer science and economics has recently been making tremendous societal impact with algorithms for infrastructure security, organ exchange, ad auctions, and resident matching adopted worldwide. My research focuses on fundamental solution concepts that are broadly defined and analytically elegant as such solutions are easy to understand, and thus have potential for quick adoption in practice. I am interested in analyzing their strengths and weaknesses in practical applications. For example, in my recent work with collaborators, we consider the problem of fairly dividing a set of goods among individuals --- the central problem in the field of computational fair division, with applications including inheritance division and divorce settlement, and show that simply maximizing the Nash welfare (product of utilities) results in a compelling fairness and efficiency guarantees not provided by any other known solution concept in the literature. Similarly, in another recent work, we study a different principled solution concept, the leximin mechanism, which maximizes the minimum utility. It is known to be perfectly suited for fairly allocating computational resources in clusters, but we show that its fairness, efficiency, and game-theoretic properties extend to a much broader domain that subsumes previously studied real-world settings from more than 10 papers in the literature. I argue that these ideas have the potential to be applicable to broader domains of real-world problems. For instance, the MNW solution might be useful for assigning rooms among roommates and dividing the rent by modelling money as a divisible good, and constraining the allocation space so that each roommate receives a room. I argue that welfare maximization can also be applied to the distant field of voting theory, in which it is typical to not ask the voters for utilities, but rather ordinal comparisons of alternatives. We recently show that aiming for the best approximation of welfare subject to ordinal information performs well on real data. But theoretical guarantees of such an approach requires further study. My work has resulted in development of two not-for-profit websites, RoboVote.org and Spliddit.org, which implement fundamental solutions in voting and fair division to solve real-world problems. The websites have attracted more than 100,000 users in three years, which shows the potential that such solutions have to help society at large.