Unformatted text preview: s still require that agents compute their values for all possible outcomes, which is often impossible. Shoham and Tennenholtz 26 consider the communication complexity of auction mechanisms. Sandholm 25 proposed Traconet, a decentralized system for dynamic task reallocation via negotiation between agents, with an application to a transport scheduling problem in which agents have hard local valuation problems. Safe and opportunistic strategies are proposed for contracting strategies with approximate valuations. Sandholm and Lesser 23 introduce decommitment to allow agents that make early mistakes because of approximate values and uncertainty to backtrack, and increase allocative e ciency. The technique allows agents to integrate local deliberation with negotiation between many other agents. The focus is on a decentralized system, while our work considers techniques for auction systems| with a centralized auctioneer. Parunak et al. 20 describe a market-based constraint system for interactive decision support between cooperative agents. The MarCon system aims to reduce agents' work; agents can express broad initial preferences and must only re ne preferences as necessary to compute a good system-wide solution. Models of economic search with one-time participation costs have demonstrated that posted-price sequential auctions can reduce agent uncertainty and improve performance when agents are uninformed about the bids of other agents 3, 10 . Rothkopf and Harstad 21 argue that auction design should be considered in the context of a marketplace, with agents that take active participation decisions. Although some auction models consider agents with one-time participation costs 6, 11 , the models cannot distinguish between sealed-bid and iterative auctions. 6 Conclusions
We have presented the results of an empirical comparison of the performance of rst-price ascending, second-price sealed and posted-price sequential auctions, for a simple model of agents with hard valuation problems. The empirical result...
View Full Document
- Fall '13