Belief Dynamics and Action Selection in Qualitative Approaches to Reasoning about Uncertainty

关于不确定环境中的推理和决策。处理这种不确定性的标准方法是定量的，通常利用概率论。这个项目的长期目标是提供更多的定性方法，可以被主体有效地用于关于不确定性的推理。该项目的重点是信念和行动之间的相互作用，以及如何在定性框架内处理这一问题。尤其令人感兴趣的是两种用途，它们对实现长期目标至关重要：信念动力和行动选择。信念动力学涉及了解代理的信念在代理进行观察或执行操作后应该如何改变。为了捕捉信念动态，使用了一个正式的信念模型，该模型显式地包含了时间。本研究的最终目的是为定性信念动力学的一般理论提供基础。行动选择涉及理解一个代理人在给定他的信念的情况下应该如何行动。通常的决策理论方法使用一个描述代理偏好的效用函数和一个描述代理信念的概率分布。但在许多情况下，完全量化不确定性是不可行的。另一种不考虑偏好的方法描述了代理通过程序选择的操作，该程序直接根据代理的信念指定代理的行为，允许直接根据我们希望它满足的规范编制计划。

Concerning reasoning and decision making in uncertain environments. The standard approaches to dealing with this uncertainty are quantitative, and typically make use of probability theory. The long-term goal of this project is provide more qualitative methods that can be used effectively by agents reasoning about uncertainty. The project focuses on the interplay between beliefs and actions, and how this can be dealt with in a qualitative framework. Of particular interest are two uses that will be crucial in attaining the long-term goal: belief dynamics and action selection. Belief dynamics involves understanding how an agent's beliefs should change after the agent makes an observation or performs an action. To capture belief dynamics, a formal model for belief is used that includes time explicitly. The ultimate goal of this research is to provide a basis for a general theory of qualitative belief dynamics. Action selection involves understanding how an agent should act, given his beliefs. The usual decision theoretic approach uses a utility function describing the agent's preferences and a probability distribution describing the agent's beliefs. But in many situations it is infeasible to completely quantify uncertainty. An alternative approach that dispenses with preferences describes the agent's choice of action by a program directly specifying the behavior of the agent in terms of his beliefs, allowing compilation of plans directly from the specification that we want it to satisfy.