点击次数：

论文类型：会议论文

发表时间：2012-06-09

收录刊物：EI、Scopus

卷号：7345 LNAI

页面范围：689-698

摘要：Cooperation among agents is critical for agents' Artificial Intelligence (AI). In multi-agent system (MAS), agents cooperate with each other for long-term return and build such partnership in most of the time. However, the partnership could be broken easily if one agent did not or refused to grant a favor to another. Will it be helpful to MAS or individual agent, if agent has controllable level of tolerance? That is the main question of this paper. In order to find an answer, we propose a cooperative strategy, "flexible reciprocal altruism model (FRAM)". In FRAM, agent has a controllable rate of tolerance and is willing to grant favors for long-term return. Agent can determine whether to grant a favor to another based on their past interactions. As a result, granting unmatched favors by accident will not break the relationship between two agents immediately. Experiments show that our strategy performs well with different cost/value tradeoffs, numbers of agents, and load. ? 2012 Springer-Verlag.