Rough set-based empirical modeling with applications to control, pattern classification adn data mining

提出的研究计划涉及粗糙集理论及其概率扩展的基础研究、算法和应用。粗糙集理论及其相关技术主要涉及从数据中自动学习分类算法。基于粗糙集的方法包括分析数据对学习目的的适用性的技术，从数据中消除无用信息的技术，同时关注最主要的分类因素，以及从数据中自动形成分类规则和决策表的技术。他说，目前的提议侧重于对粗集的概率方法，特别是可变精度和概率的非粗集模型。概率扩展允许对数据中存在的随机关系进行建模，并开发具有可控不确定性程度的分类器系统。他说，该提案的应用方面集中在三个主要应用领域。第一个是关于通过机器学习从数据中获取控制算法的理论、算法和实验研究。这一过程的关键目标是用从操作数据自动生成控制算法来代替复杂的数学建模步骤。他说，第二个应用领域涉及选定的模式识别应用的基本方法、算法和原型制作。特别是，将用医学图像解释、人脸图像识别、文本内容分类和孤立语音的非特定说话人识别等自动图像分类进行实验。他说，第三个广泛的应用领域是数据挖掘，其中基于粗糙集的算法将用于分析和建模数据中存在的概率依赖关系，检测关系中的基本因素，并优化派生模型。主要的重点将是与医学研究机构合作分析和解释医学数据。

The proposed research program is concerned with the fundamental research, algorithms and applications of rough set theory and its probabilistic extensions. The theory of rough sets and the related technologies deal in the most part with automated learning of classification algorithms from data. The rough set-based methods include techniques for the analysis of suitability of data for the learning purpose, techniques for the elimination of  useless information from data, while focusing on the most dominant classificatory factors, and   the techniques for automated formation of classification rules and decision tables from data.      The current proposal is focused on probabilistic approaches to rough sets, in particular on variable precision and probabilistic  rough set models. The probabilistic extensions allow to model stochastic relationships existing in data and to develop classifier systems operating with controlled  degree of uncertainty.      The application aspect of the proposal is focused on three major application domains. The first one is related to the theory, algorithms and experimental investigation of the problem of control algorithm acquisition from data through  machine learning. The key objective in this process is to substitute the complex mathematical modeling steps with automated generation of control algorithm from operation data.      The second application domain deals with the  methodology, algorithms and prototyping of selected pattern recognition applications. In particular, experiments will be conducted with  automated image classification such as medical image interpretation, face image recognition, text contents classification and  speaker-independent recognition of isolated spoken words.      The third broad application area is data mining in which the rough set-based algorithms will be used for analysis and modeling of probabilistic dependencies existing in data, detection of fundamental factors in the relationships and optimization of the derived models. The main focus will be on analysis and interpretation of medical data in cooperation with medical research institutions.