Discovery, representation and analysis of statistical and structural patterns

这项拟议的研究是对目前NSERC发现资助项目的扩展，该项目的标题是：“统计和结构模式的发现、表示和分析”。它的目标是：1)在存在噪声的情况下从各种类型的数据中自动发现统计和结构模式；2)开发用于分析、解释和处理的模式表示的统一框架；3)将所开发的技术应用于需要智能解决方案的现实生活问题。通过与模式发现软件系统有限公司(PDS)的合作，模式发现(PD)的理论和算法已经发展成一个称为DISCOVER*E的软件系统，在解决制药、油砂、天然气和焦炭加工、生物信息学和医疗保健等领域的复杂现实问题方面取得了巨大成功。我最近的研究重点是模式发现及其扩展的科学基础，包括：1)统计模式的聚类及其在不同数据子空间中发现它们关联的数据组；2)发现数据组的结构关系；3)连续属性的模糊划分；4)属性聚类、特征选择和重新汇集用于分类和解释；(5)分布不均匀的样本的Pd。基本工作已被记录在期刊出版物和会议记录中。在结构模式和知识表示方面，获得了美国专利。它的属性超图表示(AHR)和操作将被扩展到涵盖更广泛的数据和模型。即将到来的研究将集中在：1)各种数据子空间中统计模式和数据组的发现过程和知识表示的泛化(在AHR中)，走向更通用的大型数据库和系统的PD框架；2)发现图像中依赖于分辨率的模式，用于分割、对象/场景表示、分类和识别；3)PD在a)石油勘探、提取和精炼中的应用；b)医疗保健在诊断和预后中的应用；c)基因表达数据和d)基因组途径。

The proposed research extends the current NSERC Discovery Grant Project entitled: "Discovery, Representation and Analysis of Statistical and Structural Patterns". Its objectives are: 1) to automatically discover statistical and structural patterns from various types of data in the presence of noise; 2) to develop a unified framework of pattern representations for analysis, interpretation and manipulation and 3) to apply the developed technologies to real-life problems which require intelligent solutions. Through collaboration with Pattern Discovery Software Systems Ltd (PDS), the theories and algorithms of pattern discovery (PD) have been developed into a software system known as discover*e with great success in solving complex real world problems in areas covering pharmaceuticals, oil sand, gas and coke processing, bioinformatics and healthcare, etc. My recent NSERC research focus is on the scientific fundamentals of PD and its extensions including: 1) clustering of statistical patterns and discovery of their associated data groups in various data subspaces; 2) discovering the structural relations of data groups; 3) fuzzy partitioning for continuous attributes; 4) attribute clustering, feature selection and re-pooling for classification and interpretation; and 5) PD for samples with unbalanced distributions. Fundamental works have been documented in journal publications and conference proceedings. In structural pattern and knowledge representation, a US Patent was granted. Its attributed hypergraph representations (AHR) and manipulation will be extended to cover a broader range of data and models. The upcoming research will focus on: 1) the generalization of the discovery process and knowledge representation of statistical patterns and data groups in various data subspaces (in  AHR), moving towards a more general PD framework for large database and systems;  2) the discovery of resolution-dependent patterns in images for segmentation, object/scene representations, categorization and recognition; 3) applications of PD to a) oil exploration, extraction and refining; b) healthcare in diagnostics and prognostics; c) gene expression data and d) genomic pathways.