Unformatted text preview: Work
In Parkes and Ungar 15 we introduce a formal de nition for bounded-rational compatible BRC auctions, to describe auctions in which agents can bid optimally with approximate values. Iterative auctions are a special class of BRC auctions which can compute optimal allocations with enough agent computation. Empirical results for a model of agents with limited computation highlight the importance of BRC auctions in combinatorial allocation problems, with agents that need bundles of items. Iterative bundle auctions, such as iBundle 16, 18 , that allow agents to perform incremental value computation and adjust their bids in response to bids from other agents, are particularly important in applications to hard distributed optimization problems. There is a growing literature on auctions for e-commerce, see 7 for an introduction. Hard valuation problems are relevant in many online consumer auctions, for example for collectibles and refurbished electronic goods. This provides an appealing explanation for why many on-line auctions are ascending-price, while few on-line auctions are sealed-bid. This explanation seems more probable than alternative explanations in terms of risk-seeking agents that enjoy risk-taking in iterative auctions 13 . We believe that this work is the rst to compare the performance of auctions with a normative deliberation model for agents with limited or costly computation. Early work in market-oriented programming 29 assumed that agents were provided with closed-form solutions to their local valuation problems, so 11 minimizing agent valuation work was not important. Problems associated with limited or costly computation in the auctioneer have received recent attention, in particular with respect to the generalized Vickrey auction 27 . Interesting recent work explores methods to introduce approximate solutions but retain the incentive-compatibility property such that truth-telling remains optimal for self-interested agents 8, 9, 14 . However, these method...
View Full Document
This document was uploaded on 03/19/2014 for the course COMP CS286r at Harvard.
- Fall '13