近年來,機(jī)器學(xué)習(xí)取得了飛速的發(fā)展。由于最小二乘法在問題制定和實(shí)施上的簡單性,在過去幾年中受到了廣泛的關(guān)注。盡管最小二乘模型在分類和回歸方面具有良好的性能,但它對參數(shù)設(shè)置很敏感。這一挑戰(zhàn)使研究人員更加關(guān)注這類單一的模型方法。解決這類問題的有效途徑就是引入集成模型。本文首先概述了研究背景和最小二乘法的應(yīng)用領(lǐng)域。之后簡要討論了研究現(xiàn)狀,并介紹了最小二乘法的最后預(yù)處理方面。在目前的研究過程中,核集成學(xué)習(xí)在其應(yīng)用方面取得了顯著的進(jìn)展。這主要通過提出正則化核集成回歸、耦合最小二乘支持向量集成機(jī)和樣本誘導(dǎo)因子核集成回歸來實(shí)現(xiàn)的,主要內(nèi)容如下:1)提出了聯(lián)合正則核集成回歸方案。該方案將多個(gè)核回歸器同時(shí)應(yīng)用到一個(gè)統(tǒng)一的集成回歸框架中,并通過最小化核希爾伯特空間中的總集成損失函數(shù)來實(shí)現(xiàn)共同正則化。通過這種方式,一個(gè)對數(shù)據(jù)進(jìn)行更精確擬合的核回歸器可以自動獲得更大的權(quán)重,從而獲得更好的整體集成性能。與梯度增強(qiáng)法、回歸樹法、支持向量回歸法、嶺回歸法、隨機(jī)森林法等一些單模型和集成回歸方法相比,我們提出的方法可以在在多個(gè)UCI數(shù)據(jù)集上實(shí)現(xiàn)回歸和分類問題的最佳性能。2)提出了一種新的基于耦合最小二乘的集成支持向量機(jī)（... 

【文章來源】：江蘇大學(xué)江蘇省

【文章頁數(shù)】：133 頁

【學(xué)位級別】：博士

【文章目錄】：
DEDICATION
ABSTRACT
摘要
Chapter 1 Introduction
    1.1 Background of least squares
    1.2 Significance of the study
    1.3 Challenges in the Least Squares Problems
    1.4 Contributions of the Dissertation
    1.5 The Organization of Dissertation
Chapter 2 Related Works
    2.1 Introduction
    2.2 Regression
        2.2.1 Single Model Regression(Linear regression)
            2.2.1.1 Ridge Regression
            2.2.1.2 Lasso Regression
            2.2.1.3 ElasticNet Regression
            2.2.1.4 Linear Regression
        2.2.2 Non-linear based regression
            2.2.2.1 Kernel Ridge Regression
            2.2.2.2 Support Vector Regression
        2.2.3 Ensemble Model Regression
            2.2.3.1 Random Forest
            2.2.3.2 Gradient Boosting
            2.2.3.3 Adaboost
            2.2.3.4 Decision Tree Regression
    2.3 Classification
        2.3.1 Decision Tree
        2.3.2 Boosting
Chapter 3 Co-Regularized Kernel Ensemble Regression
    3.1 Introduction
    3.2 RKHS and the Representer Theorem
    3.3 Kernel ridge regression and ensemble model
    3.4 The proposed Method
        3.4.1 Co-regularized kernel ensemble regression
    3.5 Experimental Results
        3.5.1 Dataset description
        3.5.2 Experimental settings
        3.5.3 Performance Evaluations and comparisons
        3.5.4 Classification
            3.5.4.1 Data description
        3.5.5 Digits Recognition
    3.6 Brief Summary
Chapter 4 Coupled Least Squares Support Vector Ensemble Machines
    4.1 Introduction
    4.2 Related works in coupled idea
    4.3 The Proposed Method
        4.3.1 Coupled least squares support vector ensemble machine(C-LSSVEM)
    4.4 Experiments
    4.5 Experimental Settings
    4.6 Experimental results
        4.6.1 Artificial dataset
        4.6.2 UCI datasets
        4.6.3 Handwritten digits-datasets
        4.6.4 NWPU-RESISC45 dataset
    4.7 Brief summary
Chapter 5 Sample-Induced Factorization Kernel Ensemble Regression
    5.1 Introduction
    5.2 Sample-induced factoring idea
    5.3 The Proposed Method
        5.3.1 Sample-induced factorization kernel ensemble regression
    5.4 Experimental results
    5.5 Parameter setting
        5.5.1 Regression on UCI Dataset
        5.5.2 Classification experiments
    5.6 Classification and computer vision dataset
    5.7 Brief summary
Chapter 6 General Conclusions and Future Works
    6.1 General Conclusions
    6.2 Contributions
    6.3 Future Work
References
Acknowledgement
Publications


【參考文獻(xiàn)】：
期刊論文
[1]基于電子商務(wù)用戶行為的同義詞識別[J]. 張書娟,董喜雙,關(guān)毅.  中文信息學(xué)報(bào). 2012(03)



本文編號：3548332