Elisabeth Crawford, Manuela Veloso
We define the Extended Incremental Multiagent Agreement Problem with Preferences (EIMAPP). In EIMAPP, variables arise over time. For each variable, a set of distributed agents receives reward for agreeing on which option to assign to the variable. Each of the agents has an individual, privately owned preference function for choosing options. EIMAPPs reflect real world multiagent agreement problems, including multiagent meeting scheduling and task allocation. We analyze negotiation in EIMAPP theoretically. We introduce semi-cooperative agents, which we define as agents with an increasing incentive to reach agreements as negotiation time increases. Agents necessarily reveal information about their own preferences and constraints as they negotiate agreements. We show how agents can use this limited and noisy information to learn about other agents, and thus to negotiate more effectively. We demonstrate our results experimentally.
Subjects: 7.1 Multi-Agent Systems; 12. Machine Learning and Discovery
Submitted: May 6, 2008