正态云模型研究回顾与展望

Retrospect and Prospect of Research of Normal Cloud Model

不确定性信息的表达和处理是人工智能的一个重要研究问题.目前有多种理论模型从不同的角度研究不确定性问题,包括模糊集、粗糙集、概率论、证据理论等.1995年,李德毅院士在概率论和模糊集理论两者的基础上,提出了一种处理不确定性问题的双向认知模型——云模型,即通过正向云变换和逆向云变换算法进而实现定性概念与定量数值的双向转换.经过20多年的研究与发展,该模型逐渐得到完善,并在不确定性信息处理方面得到了广泛应用.该文回顾了正态云模型理论的研究现状和进展.并在此基础上分析了其存在的挑战和问题:(1)双向认知计算方面,虽然多步逆向云能够实现稳定的双向认知变换,但是它是基于单个云概念的,尚未有多粒度云模型双向认知变换的研究成果;(2)云模型相似性度量方面,由于不同的领域问题需要不同的评价标准,需要开展针对特定问题的云模型相似性度量研究;(3)粒计算机制方面,高斯云变换能够实现由细到粗的粒度变换以及多粒度概念的自适应生成,解决了云模型的变粒度问题.但是没有体现不同云概念以及不同粒度层次之间的关联;(4)多维云模型方面,目前在这方面的研究工作相对较少,缺乏较有效的多维云表示方法.该文针对以上问题,围绕当前的研究热点——大数据存在的挑战,进一步提出了大数据的云模型研究框架,深入探讨了未来的研究方向,并指出未来的工作需要以大数据为中心,结合粒计算、机器学习以及统计学的思想,进一步完善云模型的理论机制,该文的工作对于大数据和云模型理论的研究提供了重要的参考价值.

The expressing and processing of uncertain information is one of the key basic issue in the era of artificial intelligence. Currently, there have been many theories, such as fuzzy set, rough set, probability statistics, evidence theory, etc. , to deal with uncertainty information from different perspectives. The cloud model theory as a new cognition model for uncertainty proposed by Prof D. Y. Li in 1995 based on probability theory and fuzzy set theory, which provides a way to realize the bidirectional cognitive transformation between qualitative concept and quantitativedata by forward cloud transformation algorithm （FCT） and backward cloud transformation algorithm （BCT）. After more than 20 years of research and development, Cloud model has been extensively studied and gradually improved, and it has been widely applied in uncertain information processing in the past decades. Through reviewing the fundamental results and state-of-art research, we discuss some key problems and challenges for normal cloud model, including. （1） Although stable bidirectional cognition transformation between qualitative concept and quantitative data can be realized by multi-step backward cloud transformation, it is still only a cognition transformation based on single cloud concept, and there are still lack of results involving cognition transformation on multi-granularity cloud concepts. （2） For the similarity measure of cloud model, there exist different evaluation criterions for the similarity measure of cloud models in many fields, therefore it is necessary to develop the similarity measure of cloud model for specific application problems. （3） For granular computing mechanism of cloud model, the Gaussian Cloud Transformation （GCT） can realized the granularity transformation from fine to coarse and generate the multi- granularity concepts adaptively, which solve the problem of variable granularity. However, it has no ability to reflect the relationship between different cloud concepts or different granularity. （4） Owing to lack of studies on multi-dimensional cloud model, there are still no valid and reliable methods for its representation. Based on the above problems, in this paper, we point out the existing challenge problems in big data which is a current research issue, and we propose research framework for cloud model based big data. Future perspectives and potential research topics are also discussed. Finally, we point out the future work of cloud model should combine with granular computing, machine learning and statistics. These works are significant for the development of big data and cloud model.