This preview shows page 1. Sign up to view the full content.
Unformatted text preview: values the value of bS is between a; b ";
or bestresponse the bundle that maximizes bS , pS at those prices is S1 " then the
total valuation work performed by the auctioneer can be less than that required by agents
with the XOR bidding language. Savings of this kind can be realized within an algorithm
that makes explicit use of these types of query structures.
Let me outline some serious limitations of the bidding program model in some domains:
The speci cation problem can be as di cult as the valuation problem. In particular,
the assumption above is that a single speci cation allows an agent to compute its
75 value for all bundles. In many problems the agent might need to collect additional
information, consult human experts etc., to form a model with which to determine the
value of each bundle. As a concrete example, consider the FCC spectrum auction. For
any particular set of licenses a bidder might need to construct a new business model,
to determine its value, and this can require costly and timeconsuming information
gathering, conference calls, and modeling e orts Mil00a .
Local valuation problems might not be well formed, the agent might not be able to
provide a clear description of the method with which the value of each alternative is
determined. This is a particular concern in systems in which human experts must
be consulted to determine values.
The size of the speci cation of a problem might be too large to transmit to the
mechanism. Perhaps computing the value for a bundle requires access to a large
database of information?
Value and sensitivity of information. In a supplychain example, will IBM really
be happy to release the methods that it uses to take procurement decisions? This
information has considerable value to a competitor.
Trust. Can the agent trust the auctioneer to faithfully execute its bidding program?
There might be a role for veri cation mechanisms to enable an agent to verify the
value computation performed by a mechanism. Harkavy et al. HTK98 and Naor et
al. NPS99 provide an introduction to some ideas from secure distributed computation that can be used in auction environments. Dynamic Methods.
An alternative approach is to open up" the algorithm for computing the outcome of the
GVA, and involve agents dynamically in the computational process. It is easy to construct
examples in which it is not necessary to have complete information about agents' valuation
problems to compute and verify the outcome of the auction. A few simple examples are
described at the end of this section. A well structured dynamic method might ask agents
for just enough information to enable the mechanism to compute and verify the outcome.
A dynamic mechanism may elicit the following types of approximate information from
agents:
76  ordinal information, i.e. which bundle has highest value out of S1 , S2 and S3 ?"
 approximate information, i.e. is your value for bundle S1 greater than 100?"
 bestresponse information, i.e. which bundle do you want at prices pS ?"
 equivalenceset information, i.e. is there...
View
Full
Document
This document was uploaded on 03/19/2014 for the course COMP CS286r at Harvard.
 Fall '13
 DavidParkes

Click to edit the document details