三支决策不确定性度量与精准推理研究

In the era of big data, the study on uncertainty measure and reasoning once again becomes a hot topic in the field of intelligent decision. Three-way decisions, as a new model for the uncertainty problem solving, gains a competitive advantage in the face of inadequate information and cost-sensitive decision making. However, regarding uncertainty research, it still suffers from the deficiencies of the singleness for the factors considered, the low precision in decision making, and so on. Therefore, from the perspective of multiple angles (objects, attributes, relationships, granularities, thresholds, labels) and multiple granularity, this project intends to extend the existing three-way decision models into multi-label three-way decision models and to study the uncertainty measure and reasoning in a deep and comprehensive manner, providing new means and methods for making precision decisions based on big data. The main contents are as follows: 1) on uncertainty measure for three-way decisions, the focus is the measure of uncertainty derived from the interaction of key elements for probabilistic rough set three-way decisions; 2) on the solution to the optimal decision threshold for three-way decisions, the core is the optimization estimate of the cost functions partially unknown for decision-theoretic rough set; 3) on uncertainty reasoning for three-way decisions, the concentration is the region updating for probabilistic rough set based on flow calculation mode under the dynamic environment and the precision reasoning based on the fusion of key elements for three-way decisions; 4) on the multi-label three-way decision models and precision reasoning, the attention is mainly to the uncertainty measure with multiple labels dependence and the classification with multiple labels.

大数据时代，不确定性的度量和推理研究再次成为智能决策领域热点问题之一。三支决策是一种不确定性问题求解的新模型，其在信息不充分且决策代价敏感时尤具优势；但是其不确定性研究方面仍存在考虑因素单一、决策精度低等不足。本课题拟将现有的三支决策模型扩展成多标记三支决策模型，采用粒计算的多角度 (对象、属性、关系、粒度、阈值、标记)、多粒度思想，系统、深入地研究其不确定性度量与推理方法，为基于大数据的精准决策提供新的方法。研究内容主要包括：1)三支决策的不确定性度量，重点研究概率粗糙集三支决策关键要素相互作用的不确定性度量；2)三支决策的最优决策阈值求解，重点研究决策粗糙集代价函数部分缺失的最优化估计；3)三支决策的不确定性推理，重点研究动态环境下基于流式计算模式的概率粗糙集区域更新，以及基于三支决策关键要素融合的精准推理；4)多标记三支决策模型与精准推理，重点研究多标记依赖不确定性度量及分类。

本课题主要研究了在多标记三支决策目标下，采用粒计算的多角度多粒度的不确定性度量及推理方法。主要研究结果及进展如下：.(1)提出了三支决策的不确定性度量，在多标记数据的多标记学习问题中进行多种运用，降低多标记知识表示的不确定性，提升分类精度。.(2)提出了三支决策的阈值求解方法，定义了不同的概率方法，对决策粗糙集代价函数进行最优化估计。.(3)研究了由属性约简导致的区域变迁不确定性，提出了三支决策对于属性约简的不确定性度量方式的优化定义，并结合增量学习、协同训练等思想，深入研究了粒化工作在大数据环境下的高效计算课题。.(4)将提出的三支决策降低不确定性的思想引入实际应用中，均能提升实验效果，应用范围包括图像检索，行人重识别，行人搜索，行人追踪，用户图像分类等。.上述研究成果已在本领域重要国际期刊上发表，其中有28篇已被SCI收录。此外，出版学术著作4部，举办国际会议1次。这些工作不仅对不确定性分析和粒计算理论的发展起到积极的推动作用，且在大数据高效分析方面产生了重要影响。