一种基于影响图学习其他Agent模型方法

New way of learning models of other agent based on influence diagrams

为了实现动态环境中Agents之间的有效协作,Agent必须能够识别其他Agent的模型。用影响图作为Agent模型表示工具,给定Agent的一个初始模型和它的历史行为,在能力、优先和信念学习的基础上来构建新的模型。学习的方法是用其他Agent的历史行为作为训练集,利用神经网络学习技术来修改效用函数。

To achieve effective coordination among agents in dynamic environments, agents may have to recognize the models of other agents. This paper gives three strategies to create a new model of other agent, which is based on learning of its capabilities, preferences and beliefs and also gives an initial model and the agent＇s behavior history, using influence diagrams as a modeling representation tool. The method of preference learning attempts to modify the other agent＇s utility function by incorporating a neural network learning technique, using observed behavior history of other agent as training set.