A Faithful Mechanism for Privacy-Sensitive Distributed Constraint Satisfaction Problems

We consider a constraint satisfaction problem (CSP) in which constraints are distributed among multiple privacy-sensitive agents. Agents are self-interested (they may reveal misleading information/constraints if that increases their benefits) and privacy-sensitive (they prefer to reveal as little information as possible). For this setting, we design a multi-round negotiation-based incentive mechanism that guarantees truthful behavior of the agents, while protecting them against unreasonable leakage of information. This mechanism possesses several desirable properties, including Bayesian incentive compatibility and individual rationality. Specifically, we prove that our mechanism is faithful, meaning that no agent can benefit by deviating from his required actions in the mechanism. Therefore, the mechanism can be implemented by selfish agents themselves, with no need for a trusted party to gather the information and make the decisions centrally.