包含度理论——不确定性研究的方法学

THEORY OF INCLUDING DEGREES-METHODOLOGY OF UNCERTAINTY RESEARCH

不确定性推理是人工智能的关键问题。本文引进了包含度的概念以及包含度的生成方法,给出了包含度在专家系统中关于知识获取与推理的某些应用。 不确定性是人工智能中最活跃的研究领域,对于发展智能计算机有着重要意义。不确定性推理有定量方法、定性方法以及两者相结合的方法。对于定量方法,首先是与测度及不确定性信息相关。不同的测度与表示方法得到不同的不确定性推理,如包含有MYCIN不确定因子与主观贝叶斯推理的概率推理方法、证据推理方法、模糊推理方法、信息推理方法等。上述方法的共同特征是用概率测度、信息测度、似然测度、可能性测度、必然性测度给出假设的度量。不确定性推理的本质是在各种测度下对于包含关系给出一种估计。概括已有的不确定性推理使我们引进了包含度的概念,它足以表达各种不确定性推理的共同特征。

Uncertainty inference is a key problem for artifical intelligence. In this paper, the concept of including degree is introduced and some methods of generating including degrees are given. Some applications of including degrees to the retrieval and inference in expert systems are demonstrated. Uncertainty inference is one of the most active research fields in artifical intelligence as well as an important means for developing intelligent computers. Its main approaches are quantitative methods, qualitative methods and other methods incorporating both. For the quantitative methods approach, the first step is to represent and measure the uncertain information. Different representing and measuring methods result in different kinds of uncertainty inference. Currently, the widely used methods include probability-based inference meth- ods such as methods of MYCIN and methods of subjective Bayes, evidence theory-based inference methods, fuzzy set-based inference methods and information-based methods, etc.. The common character of the above methods is to measure the hypothesis with a measure which may be probability measure, belief measure, likelihood measure, possibility measure and necessity measure, etc. . The nature of uncertainty inference is to give evaluations on in elusion relationships by means of various measures. On the basis of abstracting the existing methods of uncertainty inferences, this paper introduces the concept of including degree which catches the common characters of various methods of uncertainty inferences.