This preview shows page 1. Sign up to view the full content.
Unformatted text preview: ossible outcomes.
The revelation principle provides a very important theoretical tool, but is not useful
in a constructive sense in di cult domains. The transformation assumed in the revelation
principle from indirect mechanisms e.g. an iterative auction to directrevelation mechanisms e.g. a sealedbid auction assumes unlimited computational resources, both for
agents in submitting valuation functions, and for the auctioneer in computing the outcome
of a mechanism Led89 . In particular, the revelation principle assumes:
 agents can compute and communicate their complete preferences
 the mechanism can compute the correct outcome with complete information about
all relevant decentralized information in the system.
It can soon become impractical for an agent to compute and communicate its complete
preferences to the mechanism, and for the mechanism to compute a solution to the centralized optimization problem. Directrevelation mechanisms convert decentralized problems
into centralized problems.
Yet, the revelation principle does have a vital role in the design of all mechanisms,
both direct and indirect, sealedbid and iterative. The revelation principle provides focus
63 to the design of iterative mechanisms. Taken along with the uniqueness of the Groves
mechanisms amongst all e cient and dominantstrategy mechanisms Section 2.4, then
any e cient and iterative mechanism with incremental truthrevelation as a dominant
strategy must compute the outcome of a Groves mechanism for the underlying preferences
of agents. This is to avoid the existence of an equivalent directrevelation mechanism for
the mechanism that is outside of the class of Groves, which would be impossible.
It is useful to characterize the computation in a mechanism within two distinct levels:
1. At the agent level:
a Valuation complexity. How much computation is required to provide preference
information within a mechanism?
b Strategic complexity. Must agents model other agents and solve gametheoretic
problems to compute an optimal strategy?
2. At the infrastructure level:
a Winnerdetermination complexity. How much computation is expected of the
mechanism infrastructure, for example to compute an outcome given information
provided by agents?
b Communication complexity. How much communication is required, between
agents and the mechanism, to compute an outcome?
Dominant strategy mechanisms, such as the Groves mechanisms, are e cient and
strategyproof mechanisms, giving them excellent strategic complexity. An agent can compute a dominantstrategy without modeling the other agents and without gametheoretic
reasoning. However, the directrevelation property of Groves mechanisms provides very
bad in fact worstcase agent valuation complexity. An optimal bidding strategy requires
that an agent determines its complete preferences over all possible outcomes. Complete
information is required in all instances, even though it is often possible to solve a particular instance with incomplete information, with an interactive solution such as that
pr...
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