Negotiation Considering Privacy Loss on Asymmetric Multi-objective Decentralized Constraint Optimization Problem

• Published in: Agents and Artificial Intelligence Vol. 13251; pp. 85 - 110
• Main Author: Matsui, Toshihiro
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2022; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Asymmetric constraint optimization; Decentralized problem solving; Multi-objective; Multiagent system; Negotiation; Privacy; Social welfare
• ISBN: 9783031101601; 303110160X
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-031-10161-8_5

Reducing the information revealed by agents is an important requirement in multiagent systems. Privacy preservation has been studied in several research areas including automated negotiating agents and Distributed Constraint Optimization Problems (DCOPs). Although one of recent approaches for privacy preservation is applying secure computation to negotiation process, publishing some information during the negotiation is necessary in the cases where agents should evaluate the reason of their agreement. We propose a negotiation framework consisting of two types of Asymmetric Multi-Objective DCOPs that are repeatedly solved to gradually publish the utility values of agents until an agreement is reached. One problem defines the selection of utility values that are gradually published by agents while another is the problem consist of the published utility values and related assignments to variables. With the general formalization of DCOPs, there are opportunities to represent several classes of negotiation problems. We define several constraints for heuristic strategies that choose the utility values to be published in the next negotiation step. The criterion of social welfare to consider the multiple objectives among individual agents are also introduced. Moreover, we apply two additional preprocessing methods to adjust the utilities among agents by transferring some amount of utility values and to reduce the complexity of the original problem by approximating several relationships of agent pairs. We experimentally evaluate the effect and influence of our proposed approaches including the heuristics to select the published utility information and the additional preprocessing methods.