發(fā)布時間：2021-05-19 23:14

　　聚類是一種重要的數(shù)據(jù)分析手段。通過聚類分析,人們能有效地發(fā)現(xiàn)隱含在數(shù)據(jù)集中的數(shù)據(jù)分布特性,從而為進一步充分、有效地利用數(shù)據(jù)奠定良好的基礎(chǔ)。隨著信息技術(shù)的迅猛發(fā)展,聚類所面臨的不僅是數(shù)據(jù)量越來越大的問題,更重要的還是數(shù)據(jù)的高維度問題。但是,受“維度效應(yīng)”的影響,許多在低維數(shù)據(jù)空間表現(xiàn)良好的聚類方法運用在高維空間上往往無法獲得好的聚類效果,這對高維數(shù)據(jù)聚類分析技術(shù)提出了很大的挑戰(zhàn)。高維數(shù)據(jù)聚類是聚類分析技術(shù)的重點和難點,基于譜聚類的子空間聚類方法是實現(xiàn)高維數(shù)據(jù)聚類的有效途徑。子空間聚類的目的是將來自不同子空間的高維數(shù)據(jù)分割到本質(zhì)上所屬的低維子空間,它是高維數(shù)據(jù)聚類的一種新方法,在機器學(xué)習(xí)、計算機視覺、圖像處理和系統(tǒng)辨識等領(lǐng)域有廣泛的應(yīng)用。本文針對高維數(shù)據(jù)的子空間聚類問題給出了 一些新的聚類模型,主要工作包括以下幾個方面:1、通過分析自表示系數(shù)矩陣與聚類指標(biāo)矩陣之間的關(guān)系,我們提出了一個新的相似度學(xué)習(xí)和子空間聚類的統(tǒng)一極小化框架——基于Direction-Grouping-Effect-Within-Cluster的結(jié)構(gòu)稀疏子空間聚類（SSDG）。在SSDG中,為了讓本質(zhì)上屬于同一子空間的數(shù)... 

【文章來源】：西安電子科技大學(xué)陜西省 211工程院校 教育部直屬院校

【文章頁數(shù)】：126 頁

【學(xué)位級別】：博士

【文章目錄】：
ABSTRACT
摘要
List of Symbols
List of Abbreviations
Chapter 1 Introduction
    1.1 Background and significance of the research
    1.2 Introduction to concepts and technologies related to clustering algorithms
        1.2.1 The conventional clustering algorithm
        1.2.2 The subspace clustering
    1.3 Notations
    1.4 The main contributions of the dissertation
Chapter 2 Structured Sparse Subspace Clustering with Direction-Grouping-Effect-Within-Cluster
    2.1 Introduction
    2.2 Related work
    2.3 A unified optimization framework for Structured Sparse Subspace Cluster-ing with DGEWC
    2.4 An alternating minimization algorithm for our model
        2.4.1 Solution of the representation coefficient
        2.4.2 Spectral clustering
        2.4.3 Summary of the proposed algorithm
    2.5 Experiments
        2.5.1 Experiments on Extended Yale B dataset
        2.5.2 Experiments on COIL20 dataset
        2.5.3 Experiments on USPS dataset
        2.5.4 Experiments on ORL dataset
        2.5.5 Experiments on Hopkins 155 dataset
        2.5.6 Runtime costs
        2.5.7 Convergence analysis
    2.6 Conclusion
Chapter 3 Discriminative and Coherent Subspace Clustering
    3.1 Introduction
    3.2 Discriminative and Coherent Subspace Clustering
        3.2.1 Motivation
        3.2.2 DCSC model
        3.2.3 Minimization algorithm
        3.2.4 Connections and differences between DCSC and other related methods
        3.2.5 Summary of the proposed algorithm
    3.3 Experiments
        3.3.1 Experiments on Extended Yale B dataset
        3.3.2 Experiments on the USPS dataset
        3.3.3 Experiments on the Hopkins 155 dataset
        3.3.4 Convergence analysis
    3.4 Conclusion
Chapter 4 Discrimination Enhanced Spectral Clustering
    4.1 Introduction
    4.2 Related work
    4.3 Discrimination Enhanced Spectral Clustering
        4.3.1 Motivation
        4.3.2 DESC model
    4.4 Minimization algorithm
        4.4.1 Weighted sparse spectral clustering
        4.4.2 Solution of the representation coefficient
        4.4.3 Summary of the proposed algorithm
    4.5 Experiments
        4.5.1 Experiments on Extended Yale B dataset
        4.5.2 Experiments on the USPS dataset
        4.5.3 Convergence analysis
    4.6 Conclusion
Chapter 5 Block Diagonal Spectral Clustering
    5.1 Introduction
    5.2 Related work
    5.3 Block Diagonal Spectral Clustering
    5.4 Minimization algorithm
        5.4.1 Minimization algorithm of BDSpeCl
        5.4.2 Minimization algorithm of BDSpeC2
    5.5 Experiments
        5.5.1 Experiments on the Hopkins 155 dataset
        5.5.2 Experiments on the MNIST dataset, USPS dataset and PIE dataset
        5.5.3 Convergence analysis
    5.6 Conclusion
Chapter 6 Conclusions and Future Work
References
Acknowledgements
Curriculum Vitae


【參考文獻】：
期刊論文
[1]FDBSCAN:一種快速 DBSCAN算法（英文）[J]. 周水庚,周傲英,金文,范曄,錢衛(wèi)寧.  軟件學(xué)報. 2000(06)



本文編號：3196609