本文關(guān)鍵詞：大數(shù)據(jù)下的數(shù)據(jù)選擇與學(xué)習(xí)算法研究　出處：《西安電子科技大學(xué)》2015年博士論文　論文類型：學(xué)位論文

  更多相關(guān)文章： 主動(dòng)學(xué)習(xí) 遷移學(xué)習(xí) 和范數(shù) 黎曼商流形 非線性共軛梯度 固定低秩矩陣填充

【摘要】：信息爆炸時(shí)代給我們帶來(lái)了無(wú)論種類還是數(shù)量上都空前巨大的信息。隨著計(jì)算機(jī)通信與互聯(lián)網(wǎng)技術(shù)、各種傳感器所帶來(lái)的物聯(lián)網(wǎng)技術(shù)的極速發(fā)展與廣泛應(yīng)用,大量數(shù)據(jù)的收集變得非常容易且成本低廉。這為人工智能領(lǐng)域中迫切需求的機(jī)器學(xué)習(xí)、模式識(shí)別與計(jì)算機(jī)視覺的快速發(fā)展提供了必要的數(shù)據(jù)支撐。然而,如何有效地選擇數(shù)據(jù),如何從數(shù)據(jù)中學(xué)習(xí)有用的信息,成為擺在科研人員面前的重要問題。本文圍繞數(shù)據(jù)選擇和數(shù)據(jù)內(nèi)在子空間和流形信息學(xué)習(xí)等問題通過模型建立、算法設(shè)計(jì)和分析等方面進(jìn)行了系統(tǒng)性的研究,并將相關(guān)算法應(yīng)用于協(xié)同過濾、圖像修補(bǔ)和視頻背景建模等工程領(lǐng)域。本論文的研究成果有:1.針對(duì)海量數(shù)據(jù)的人工標(biāo)記需要花費(fèi)高昂的人力和時(shí)間成本,主動(dòng)學(xué)習(xí)作為一種適宜的最小化標(biāo)記成本的方法被越來(lái)越多的研究者所關(guān)注。在已有的主動(dòng)學(xué)習(xí)算法中,有的方法利用了未標(biāo)記數(shù)據(jù)的結(jié)構(gòu)信息,但代表數(shù)據(jù)點(diǎn)的選擇需要額外的計(jì)算,例如層次聚類;有的方法需要每次迭代預(yù)先訓(xùn)練多個(gè)分類器,從集成的角度找出需要人工標(biāo)記的數(shù)據(jù);有的方法僅僅考慮每次迭代中最靠近最優(yōu)決策面的數(shù)據(jù)。為了克服上面的不足,我們提出了一種成對(duì)K近鄰偽剪輯的主動(dòng)學(xué)習(xí)算法。該方法受K近鄰剪輯預(yù)處理思想的啟發(fā),并且在每次迭代中僅需要訓(xùn)練一個(gè)分類器和考慮最優(yōu)分類超平面附近的多個(gè)數(shù)據(jù)。同時(shí),我們也給出了相應(yīng)的算法復(fù)雜度分析和參數(shù)分析。大量的實(shí)驗(yàn)結(jié)果表明了本章提出的成對(duì)K近鄰偽剪輯的主動(dòng)學(xué)習(xí)算法相對(duì)于其他主流的主動(dòng)學(xué)習(xí)算法在僅需查詢并標(biāo)記少量樣本下就能獲得較好的分類性能。2.低秩矩陣填充與恢復(fù)問題是典型的從已知數(shù)據(jù)中學(xué)習(xí)其內(nèi)在結(jié)構(gòu)和信息的實(shí)際問題。最近幾年,這個(gè)問題在數(shù)據(jù)池環(huán)境中通過矩陣的跡范數(shù)最小化技術(shù)或其他奇異值分解的變種方法得到了很好的解決。在這種環(huán)境中,海量數(shù)據(jù)的規(guī)模、樣本的大小和視頻幀數(shù)等都是提前獲得的。所以前面的問題能夠通過在每次迭代中對(duì)數(shù)據(jù)(稀疏)矩陣進(jìn)行奇異值分解來(lái)解決,但時(shí)間復(fù)雜度非常高,因此這類方法并不適合應(yīng)用于實(shí)時(shí)的環(huán)境中。為了能實(shí)時(shí)的對(duì)視頻流進(jìn)行背景建模,本文提出了一種-范數(shù)框架下基于Grassmannian流形的在線梯度下降算法模型。應(yīng)用該模型,能在數(shù)據(jù)流的環(huán)境中在線的解決矩陣填充與恢復(fù)問題。通過引入黎曼流形優(yōu)化,沿著Grassmannian流形測(cè)地線的最優(yōu)子空間能夠被找到。作為增量學(xué)習(xí),在每次迭代中只涉及一個(gè)數(shù)據(jù)樣本(向量)的計(jì)算。-范數(shù)框架的設(shè)計(jì)是為了能從被稀疏大噪聲(局外值)和高斯噪聲污染的數(shù)據(jù)中逼近恢復(fù)原始數(shù)據(jù)�；诔俗咏惶娣较蚍ê蚲rassmannian流形優(yōu)化的一種迭代算法被提出以解決在線環(huán)境下的魯棒低秩矩陣填充、魯棒低秩矩陣恢復(fù)以及視頻監(jiān)控中的背景建模等問題。此外,一種新穎的自適應(yīng)步長(zhǎng)策略被提出來(lái)有效地追蹤子空間的變化。大量的人工和實(shí)際數(shù)據(jù)的實(shí)驗(yàn)表明,本文的方法與其他主流的算法相比擁有更好的魯棒性和有效性。3.從已知數(shù)據(jù)中學(xué)習(xí)其內(nèi)在的子空間信息可以被推廣到學(xué)習(xí)其滿秩矩陣分解背后的黎曼商流形結(jié)構(gòu),其中低秩約束可以通過滿秩矩陣分解來(lái)表示。為了能解決更一般的矩陣填充問題,這其中包括病態(tài)矩陣和大規(guī)模矩陣,本文從測(cè)度的角度分析了現(xiàn)有的主流黎曼流形優(yōu)化算法,并首次根據(jù)黎曼幾何結(jié)構(gòu)和目標(biāo)函數(shù)的尺度信息在黎曼商流形切空間的水平子空間上構(gòu)造一種新穎的黎曼測(cè)度。在黎曼商流形上優(yōu)化所需的必要組件被重新設(shè)計(jì)和計(jì)算。為了驗(yàn)證所構(gòu)造的黎曼測(cè)度的有效性,在黎曼商流形上的非線性共軛梯度法被采用。大量的數(shù)值實(shí)驗(yàn)表明,通過比較算法的收斂性,本文提出的黎曼測(cè)度優(yōu)于現(xiàn)有的黎曼測(cè)度。采用這種新穎黎曼測(cè)度的非線性共軛梯度算法在收斂性上優(yōu)于主流的低秩矩陣填充算法。4.通過結(jié)合多個(gè)個(gè)體分類器來(lái)改善單個(gè)分類器的性能近幾年越來(lái)越成為一個(gè)研究熱點(diǎn)。隨之而來(lái)的問題就是在產(chǎn)生的眾多個(gè)體分類器中是否都對(duì)降低集成系統(tǒng)的泛化誤差有益。平衡個(gè)體分類器之間的差異和個(gè)體分類器自身的準(zhǔn)確率,這本身就是設(shè)計(jì)集成學(xué)習(xí)算法的出發(fā)點(diǎn)同時(shí)也是難點(diǎn)。因此,本文提出了一種基于整數(shù)矩陣分解的選擇集成算法。該算法分別從差異性和準(zhǔn)確率兩個(gè)因素出發(fā),為了增加個(gè)體分類器之間的差異,將個(gè)體分類器的預(yù)測(cè)標(biāo)記作為原始目標(biāo),且將正確標(biāo)記引入,以此構(gòu)造一個(gè)代表個(gè)體分類器的整數(shù)矩陣,通過對(duì)該矩陣進(jìn)行分解獲得個(gè)體分類器的投影方向,最終獲得新的個(gè)體。然而,為了保證變換個(gè)體的性能,采用標(biāo)準(zhǔn)的性能判別準(zhǔn)則去除集成中性能較差的個(gè)體。最后,通過雷達(dá)一維距離像的實(shí)驗(yàn)結(jié)果表明該算法有效地平衡了個(gè)體間差異性和個(gè)體自身的準(zhǔn)確率這兩個(gè)因素,相比單個(gè)分類器和其他集成方法,該方法提高了對(duì)雷達(dá)目標(biāo)的識(shí)別準(zhǔn)確率。5.針對(duì)在一個(gè)有監(jiān)督學(xué)習(xí)任務(wù)中,如果目標(biāo)域訓(xùn)練樣本的數(shù)量非常稀少,這勢(shì)必產(chǎn)生影響目標(biāo)域中分類器學(xué)習(xí)和推廣性能的問題。為了解決這個(gè)問題,除了使用主動(dòng)學(xué)習(xí)的方法從目標(biāo)域選擇富含信息的樣本并給與標(biāo)記以增大訓(xùn)練樣本外,在某些真實(shí)環(huán)境中往往已經(jīng)存在另一些有標(biāo)記的樣本,且其獲取相比目標(biāo)域的訓(xùn)練樣本更加容易,但是這些樣本卻與目標(biāo)域的樣本具有不同的數(shù)據(jù)分布形式,這些具有不同分布的有標(biāo)記樣本構(gòu)成源域。因此,遷移學(xué)習(xí)被引入來(lái)處理目標(biāo)域訓(xùn)練樣本稀少的這類分類問題。我們提出了兩種新的遷移學(xué)習(xí)算法:第一種是基于旋轉(zhuǎn)森林空間變換的遷移學(xué)習(xí)算法,該算法通過旋轉(zhuǎn)森林空間變換將源域樣本向目標(biāo)域形成的空間進(jìn)行投影,通過測(cè)量變換后源域樣本和目標(biāo)域樣本的相似度來(lái)選擇可利用的源域樣本幫助目標(biāo)域中分類器的學(xué)習(xí)。通過文本數(shù)據(jù)的分類實(shí)驗(yàn)表明,該章所提算法相比其他算法獲得了更好的分類性能。第二種為基于數(shù)據(jù)驅(qū)動(dòng)的線性空間映射遷移集成算法。在該算法中,通過將源域的樣本向目標(biāo)域中容易被錯(cuò)分的樣本空間進(jìn)行投影變換,從而選擇出對(duì)目標(biāo)域分類有幫助的樣本加入到目標(biāo)域,改善其分類性能。特別地,為了更加有效地選擇源域樣本,本文將源域樣本進(jìn)行隨機(jī)劃分,并分別對(duì)于每個(gè)子集進(jìn)行投影變換,然后結(jié)合每個(gè)子集獲得的結(jié)果。對(duì)于UCI數(shù)據(jù)和合成孔徑雷達(dá)目標(biāo)圖像數(shù)據(jù)的分類實(shí)驗(yàn)表明本章提出的算法相比其他算法有效地提高了目標(biāo)域的分類性能,且改善了單個(gè)遷移的不穩(wěn)定性。
[Abstract]:The era of information explosion brought regardless of the type or quantity are unprecedented information for us. With the development of computer communication and Internet technology, the rapid development and wide application of Internet technology brings a variety of sensors, a large collection of data becomes very easy and low cost. This study is the urgent needs in the field of artificial intelligence machine, provide the necessary data to support the rapid development of computer vision and pattern recognition. However, how to choose the data effectively, how to learn the useful information from the data, has become an important issue in the research workers. This paper focuses on the data selection and data subspace and intrinsic information manifold learning problem through the model of system the algorithm design and analysis, and the application of the relevant algorithm in collaborative filtering, image inpainting and video background modeling engineering. Domain. The research results of this thesis are: 1. for mass data manual marking takes time and manpower cost, active learning as a method of minimizing the cost of suitable markers is concerned by more and more researchers. In the existing active learning algorithm, a method of using unlabeled data structure but, on behalf of data point selection requires additional computation, such as hierarchical clustering; some methods need each iteration pre training multiple classifiers, identify artificial markers data from the point of view of integration; some methods only consider each iteration closest to the optimal decision surface data. In order to overcome the shortcomings above, we propose an active learning algorithm a pair of clips. The pseudo K nearest neighbor method K nearest neighbor heuristic clip pretreatment thought, and only need to train a classifier and in each iteration A plurality of data considering the optimal hyperplane nearby. At the same time, we also give the corresponding algorithm complexity analysis and parameter analysis. Experimental results demonstrate that the pairwise K nearest neighbor pseudo clips of the proposed active learning algorithm with respect to other mainstream active learning algorithm only needs to query and mark can obtain a small sample the classification performance of.2. low rank matrix recovery is better filled with typical examples from the known data to study its internal structure and information. In recent years, this problem by trace norm minimization technique of matrix singular value decomposition method or other variants was solved in the data pool in this environment. In the environment of massive data, the size of the sample size and the video frames are obtained in advance. So in front of the problem can be passed in each iteration of the data (sparse) Matrix singular value decomposition to solve, but the time complexity is very high, so this kind of method is not suitable for real-time environment. In order to real-time video stream on the background modeling, this paper proposes a framework of Grassmannian - norm online gradient descent algorithm based on manifold model. This model is used to solve the matrix can online in the data stream environment in filling and recovery. By introducing the Riemann manifold optimization, along Grassmannian manifold geodesic optimal subspace can be found. As incremental learning, only a data sample involved in each iteration (vector) design calculation. - norm framework was to be from large sparse (outside noise value) and the Gauss noise pollution data approach to recover the original data. By alternating direction method and Grassmannian manifold optimization of an iterative algorithm is proposed to solution based on Is the low rank matrix robust online environment filling, the problem of robust low rank matrix recovery and video monitoring in background modeling. In addition, a novel adaptive step strategy is proposed to effectively change tracking subspace. The artificial and real data show that a large number of experiments, this method with other algorithms compared with better robustness and effectiveness of.3. from known data to study its internal space information can be extended to the Riemann manifold structure learning the full rank decomposition of matrix behind, the low rank constraint can be represented by full rank matrix decomposition. In order to solve the more general problem of filling matrix, which including the ill conditioned matrix and mass matrix, this paper analyzes the mainstream Riemann manifold existing optimization algorithms from the angle of measure, and for the first time according to the Riemann scale information geometry and objective function A novel Riemann measure constructed in Riemann flow shape tangent space level subspace. Optimizing the necessary components required in Riemann manifolds are re designed and calculated. The validity of the Riemann measure in order to verify the structure of the nonlinear conjugate gradient method in Riemann manifold is adopted. Numerical experiments show that a large number of the convergence of the algorithm, by comparison, the Riemann measure is superior to the existing Riemann measure. The performance of this novel nonlinear conjugate gradient algorithm of the Riemann measure of the convergence of low rank matrix is better than that of the mainstream.4. filling algorithm by combining a plurality of individual classifiers to improve single classifier in recent years has become a more and more the focus of research. The problem is in many individual classifier produced is beneficial to reduce the generalization error of integrated system. The balance between individual classifiers The accuracy of individual differences and the classifier itself, the starting point itself is the design of integrated learning algorithm is also difficult. Therefore, this paper proposes an integrated algorithm for integer matrix decomposition based selection. The algorithm separately from the difference and accuracy of two elements, in order to increase the difference between individual classifiers, forecast mark the individual classifier as the original target, and will be marked correctly introduced for constructing a representative individual classifier based on the integer matrix, the matrix decomposition of projection direction to obtain the individual classifier, finally get the new individual. However, in order to ensure the performance of transformation of individuals, using standard criteria for the removal performance of integrated performance is poor individual. Finally, shows that the algorithm can effectively balance the difference between individual and individual through accurate radar range profile of the experimental results The rate of these two factors, compared with the single classifier and other integration methods, this method improves the recognition accuracy of the radar target.5. in a supervised learning task, if the number of training samples of the target domain is very scarce, it is bound to have an impact in the target domain classifier learning and generalization performance. In order to solve this problem. In addition, the use of active learning methods from the target domain selection information rich samples and give marks to increase training samples, in some real environment may exist in some labeled samples, and the obtained compared to the target domain training samples more easily, but these are the sample and target domain samples with different distribution of the data, which have different distribution of labeled samples to form the source domain. Therefore, transfer learning is introduced to deal with the target domain training sample rare such Class problem. We propose two new algorithms of transfer learning: the first is the learning algorithm based on the spatial migration of rotation forest transform, the algorithm through the space rotation forest transform the source domain to the target domain formation sample space projection, similarity of source domain and target domain sample samples by measuring the transformation to select the source domain the sample can be used to help the target domain classifier learning. Through the experiment of text data classification show that the proposed algorithm has better classification performance than other algorithms. For second kinds of linear space mapping algorithm based on integrated data driven migration. In this algorithm, the source domain to the target domain in the sample easy to be misclassified sample space projection transformation, and find out the target domain classification help sample is added to the target domain, improve the classification performance. In particular, in order to more effectively Select the source domain sample, the source domain samples were randomly divided, and separately for each subset of projection transformation, and then combined with the results obtained for each subset. The experimental data of UCI and synthetic aperture radar target image data classification show that the algorithm proposed in this chapter compared to other algorithms can effectively improve the classification performance of the target domain and, to improve the individual migration instability.

【學(xué)位授予單位】：西安電子科技大學(xué)
【學(xué)位級(jí)別】：博士
【學(xué)位授予年份】：2015
【分類號(hào)】：TP181

【共引文獻(xiàn)】

相關(guān)期刊論文 前10條

1 平博;蘇奮振;周成虎;高義;;局部SVT算法的遙感反演場(chǎng)數(shù)據(jù)恢復(fù)實(shí)驗(yàn)分析[J];地球信息科學(xué)學(xué)報(bào);2011年05期

2 史加榮;焦李成;尚凡華;;不完全非負(fù)矩陣分解的加速算法[J];電子學(xué)報(bào);2011年02期

3 林杰;石光明;董偉生;;基于信息自由度采樣的信號(hào)重構(gòu)方法研究進(jìn)展[J];電子學(xué)報(bào);2012年08期

4 張芬;張成;程鴻;沈川;韋穗;;基于矩陣填充的相位檢索[J];光學(xué)學(xué)報(bào);2013年07期

5 李二俊;劉萬(wàn)林;余濤;謝東海;蔡慶空;;基于SURF算法的無(wú)人機(jī)航空?qǐng)D像自動(dòng)配準(zhǔn)研究[J];工程勘察;2013年10期

6 楊兵兵;胡士強(qiáng);;隨機(jī)抽樣一致消除特征錯(cuò)配的一種加速算法[J];電氣自動(dòng)化;2013年06期

7 李正浩;曾智洪;曾曉贏;史振寧;付仕清;;農(nóng)村信息化建設(shè)中多媒體數(shù)據(jù)的并行管理框架設(shè)計(jì)[J];重慶大學(xué)學(xué)報(bào);2013年12期

8 馬超;趙西安;王青松;;基于均勻特征匹配的無(wú)人機(jī)影像拼接[J];北京建筑工程學(xué)院學(xué)報(bào);2013年04期

9 賈豐蔓;康志忠;于鵬;;影像同名點(diǎn)匹配的SIFT算法與貝葉斯抽樣一致性檢驗(yàn)[J];測(cè)繪學(xué)報(bào);2013年06期

10 貢力;余濤;;壓縮感知在隧道病害識(shí)別中的應(yīng)用研究[J];城市道橋與防洪;2013年10期

相關(guān)會(huì)議論文 前10條

1 崔永超;李秀娟;文成林;;基于凸優(yōu)化方法對(duì)逆問題求解測(cè)量值數(shù)目的研究[A];第25屆中國(guó)控制與決策會(huì)議論文集[C];2013年

2 王亞偉;許廷發(fā);王吉暉;;改進(jìn)的匹配點(diǎn)提純算法mRANSAC[A];2013年中國(guó)智能自動(dòng)化學(xué)術(shù)會(huì)議論文集（第三分冊(cè)）[C];2013年

3 楊鴻;錢X;戴先中;馬旭東;房芳;;基于Kinect傳感器的移動(dòng)機(jī)器人室內(nèi)環(huán)境三維地圖創(chuàng)建[A];2013年中國(guó)智能自動(dòng)化學(xué)術(shù)會(huì)議論文集（第三分冊(cè)）[C];2013年

4 Biao Zhang;Qixin Cao;;3D Point Cloud Based Hybrid Maps Reconstruction for Indoor Environments[A];2013年中國(guó)智能自動(dòng)化學(xué)術(shù)會(huì)議論文集（第二分冊(cè)）[C];2013年

5 Tie Jiang;Guibin Zhu;;Image Mosaic Based on SURF and Results Optimization[A];2013年中國(guó)智能自動(dòng)化學(xué)術(shù)會(huì)議論文集（第二分冊(cè)）[C];2013年

6 張官亮;鄒煥新;孫浩;劉志波;;基于馬氏距離加權(quán)圖轉(zhuǎn)換的點(diǎn)模式匹配[A];2013年中國(guó)智能自動(dòng)化學(xué)術(shù)會(huì)議論文集（第五分冊(cè)）[C];2013年

7 Hao Shen;Chengfei Zhu;Shuxiao Li;Hongxing Chang;;A Machine Vision System for Bearing Greasing Procedure[A];2013年中國(guó)智能自動(dòng)化學(xué)術(shù)會(huì)議論文集（第一分冊(cè)）[C];2013年

8 陳明;王樹鵬;云曉春;吳廣君;;基于二維云模型過濾的重復(fù)圖像發(fā)現(xiàn)[A];2013年全國(guó)通信軟件學(xué)術(shù)會(huì)議論文集[C];2013年

9 ZHANG ChengHao;CHEN JiaBin;SONG ChunLei;XU JianHua;;An UAV Navigation Aided with Computer Vision[A];第26屆中國(guó)控制與決策會(huì)議論文集[C];2014年

10 林福良;劉海峰;張超;黃可嘉;;一種穩(wěn)定高精度的圖像特征點(diǎn)迭代匹配算法[A];第三屆中國(guó)指揮控制大會(huì)論文集（下冊(cè)）[C];2015年

相關(guān)博士學(xué)位論文 前10條

1 章寒;單倍型的分布估計(jì)和關(guān)聯(lián)分析[D];中國(guó)科學(xué)技術(shù)大學(xué);2011年

2 田彥;基于視頻的人體姿勢(shì)預(yù)測(cè)與跟蹤[D];北京郵電大學(xué);2011年

3 程捷;無(wú)線傳感器網(wǎng)絡(luò)查詢技術(shù)研究[D];華中科技大學(xué);2011年

4 劉新武;基于偏微分方程的圖像復(fù)原技術(shù)研究[D];湖南大學(xué);2011年

5 龐志峰;圖像去噪問題中的幾類非光滑數(shù)值方法[D];湖南大學(xué);2010年

6 羅自炎;Lyapunov-type對(duì)稱錐規(guī)劃[D];北京交通大學(xué);2010年

7 陳娜;矩陣恢復(fù)算法及誤差分析[D];華中科技大學(xué);2012年

8 申遠(yuǎn);一些求解結(jié)構(gòu)型優(yōu)化的一階算法[D];南京大學(xué);2012年

9 蘇雅茹;高維數(shù)據(jù)的維數(shù)約簡(jiǎn)算法研究[D];中國(guó)科學(xué)技術(shù)大學(xué);2012年

10 尚凡華;基于低秩結(jié)構(gòu)學(xué)習(xí)數(shù)據(jù)表示[D];西安電子科技大學(xué);2012年

相關(guān)碩士學(xué)位論文 前10條

1 姚璐;融合社會(huì)化標(biāo)簽的協(xié)同過濾算法研究[D];浙江大學(xué);2011年

2 賈亮;基于矩陣稀疏的視頻目標(biāo)跟蹤[D];大連理工大學(xué);2011年

3 李寅;基于張量分解的視覺顯著性算法研究[D];上海交通大學(xué);2011年

4 朱顥;全方位舌像特征提取及多核學(xué)習(xí)分類[D];哈爾濱工業(yè)大學(xué);2011年

5 郭海亮;航拍圖像增強(qiáng)處理與拼接技術(shù)實(shí)現(xiàn)[D];大連理工大學(xué);2011年

6 張慧;Bregman迭代方法及其在稀疏問題中的應(yīng)用[D];國(guó)防科學(xué)技術(shù)大學(xué);2009年

7 封婷;圖像序列射影重建技術(shù)的研究[D];南京大學(xué);2012年

8 靳正芬;求解矩陣核范數(shù)極小化問題的交替方向法[D];河南大學(xué);2012年

9 鄭錦湖;基于一種魯棒主元分析及其在目標(biāo)檢測(cè)中的應(yīng)用研究[D];云南大學(xué);2012年

10 張利慶;基于定位的移動(dòng)廣告?zhèn)�性化推薦系統(tǒng)研究[D];湖南師范大學(xué);2012年

，

本文編號(hào)：1372378