本文選題：度量學(xué)習(xí) + 魯棒學(xué)習(xí)��； 參考：《浙江大學(xué)》2017年博士論文

【摘要】：分類和排序?qū)W習(xí)是基于數(shù)據(jù)模式判別的有監(jiān)督學(xué)習(xí)問題。針對一般有監(jiān)督學(xué)習(xí)應(yīng)用的本質(zhì)需求,分類和排序?qū)W習(xí)需要考慮如下幾個因素。首先,數(shù)據(jù)在特征空間的分布中通常具有復(fù)雜的非線性幾何結(jié)構(gòu)。由于空間中的幾何拓?fù)淇赏ㄟ^一個距離度量函數(shù)刻畫,我們需要學(xué)習(xí)一般的非線性距離度量函數(shù)以有效恢復(fù)數(shù)據(jù)分布的幾何結(jié)構(gòu)。其次,數(shù)據(jù)來源中普遍存在不可靠的噪聲樣本。模型需要區(qū)分特征中可靠的模式,魯棒地反映數(shù)據(jù)的全局分布。第三,快速涌現(xiàn)的新領(lǐng)域中常常缺乏標(biāo)定的訓(xùn)練數(shù)據(jù)。這要求模型能處理訓(xùn)練和目標(biāo)數(shù)據(jù)具有相關(guān)但不同的特征和語義分布的情況,從訓(xùn)練數(shù)據(jù)中提取可適應(yīng)目標(biāo)數(shù)據(jù)的知識,實(shí)現(xiàn)有效的跨領(lǐng)域知識遷移。根據(jù)上述分析,本文從以下三個方向開展研究:挖掘數(shù)據(jù)分布的內(nèi)在幾何結(jié)構(gòu),建立學(xué)習(xí)模型的魯棒性,和實(shí)現(xiàn)可適應(yīng)的知識遷移。本質(zhì)上,這三個方面分別針對了機(jī)器學(xué)習(xí)研究的三個基本原則,即模式挖掘的有效性、魯棒性和可適應(yīng)性。它們之間具有互相促進(jìn)和補(bǔ)充的潛在聯(lián)系。有效性關(guān)心模型擬合數(shù)據(jù)分布的精確性,關(guān)注捕捉數(shù)據(jù)分布豐富的局部非線性結(jié)構(gòu)。魯棒性關(guān)心模型擬合數(shù)據(jù)分布的全局性,關(guān)注捕捉數(shù)據(jù)的整體、全局的分布結(jié)構(gòu)�？蛇m應(yīng)性關(guān)心模型提取領(lǐng)域間共享知識的能力,關(guān)注模型對未知領(lǐng)域的探索。在調(diào)研大量前人工作的基礎(chǔ)上,本論文利用上述三個研究方向間的內(nèi)在聯(lián)系,提出了新穎的排序和分類學(xué)習(xí)模型,旨在實(shí)現(xiàn)對數(shù)據(jù)模式結(jié)構(gòu)的有效挖掘,魯棒學(xué)習(xí)和知識遷移的聯(lián)合優(yōu)化。本文工作取得了如下的研究成果:一,本文提出了基于Bregman距離度量函數(shù)的結(jié)構(gòu)排序?qū)W習(xí)算法。Bregman距離函數(shù)族是一類有靈活、泛化的非線性形式的距離度量函數(shù)。算法利用該距離函數(shù)的非線性建模能力,學(xué)習(xí)適應(yīng)數(shù)據(jù)的Bregman距離函數(shù)以有效挖掘數(shù)據(jù)分布的一般結(jié)構(gòu)及語義校準(zhǔn)。另一方面,算法基于結(jié)構(gòu)學(xué)習(xí)思想建模,利用排序列表的序列結(jié)構(gòu)信息以學(xué)習(xí)適應(yīng)排序任務(wù)的模型。該算法提供了一個聯(lián)合的距離函數(shù)學(xué)習(xí)和排序?qū)W習(xí)的一般框架,通過同時建模數(shù)據(jù)的非線性分布模式和輸出列表的整體結(jié)構(gòu),實(shí)現(xiàn)了對排序模型的數(shù)據(jù)適應(yīng)和任務(wù)適應(yīng)的聯(lián)合優(yōu)化。二,本文提出了一個有效魯棒的統(tǒng)一的分類學(xué)習(xí)框架,自步提升學(xué)習(xí)(Self-Paced Boost Learning)。該框架揭示和利用了基于有效模型選擇的提升學(xué)習(xí)方法和基于魯棒樣本選擇的自步學(xué)習(xí)方法的一致性和互補(bǔ)性,將分類模型形式化為聯(lián)合的判別性模型選擇和魯棒性樣本選擇過程。模型通過同時從弱到強(qiáng)地學(xué)習(xí)分類器和從易到難地學(xué)習(xí)樣本,能夠在捕捉類間判別性模式的同時保證被學(xué)習(xí)樣本的可靠性,實(shí)現(xiàn)了分類學(xué)習(xí)的有效性和魯棒性的聯(lián)合增強(qiáng)。三,本文提出了受語義相關(guān)性約束的可適應(yīng)零樣本分類模型,從知識遷移的角度研究零樣本學(xué)習(xí)問題(從有訓(xùn)練數(shù)據(jù)的可見類別學(xué)習(xí)對無訓(xùn)練數(shù)據(jù)的目標(biāo)類別的分類器)。模型引入了新穎的語義相關(guān)性正則化(Semantic Correlation Regularization,SCR)方法,通過約束分類器的輸出符合類別間的語義相關(guān)關(guān)系,來挖掘訓(xùn)練類別和目標(biāo)類別共同的特征和語義模式,增強(qiáng)模型對目標(biāo)類別數(shù)據(jù)的適應(yīng)性。模型基于聯(lián)合的受SCR正則化的提升優(yōu)化和自控制的樣本選擇作形式化,通過對分類器的判別性、魯棒性和跨語義可適應(yīng)性的聯(lián)合增強(qiáng),在零樣本學(xué)習(xí)上實(shí)現(xiàn)了有效的知識遷移。
[Abstract]:Classification and ordination of learning is based on the data in the pattern recognition of supervised learning problems. According to the general supervised learning essence demand application, classification and ordination of learning need to consider several factors. Firstly, the distribution of data in the feature space usually has complicated structure. Because of nonlinear geometric geometry in topological spaces by a distance metric to describe, we need to learn the general nonlinear distance measure function to effectively restore the geometric structure of the data distribution. Secondly, noise samples unreliable exists in the source data. The model needs to distinguish between reliable patterns, robust reflect the distribution of the global data. Third, the training data calibration often lack the rapidly emerging new areas in this model can handle the training requirements. And the target data is related but different features and semantic distribution from the training data The extraction can be adapted to the target data to achieve cross domain knowledge, effective knowledge transfer. Based on the above analysis, this paper carried out the research from the following three aspects: the intrinsic geometric structure of distributed data mining, establish a robust learning model, and realize knowledge transfer can adapt. In essence, these three aspects respectively for three the basic principles of machine learning, namely the effectiveness of pattern mining, robustness and adaptability. With mutual promotion and complement the potential links between them. The effectiveness of care for accuracy of the model fitting the data distribution, attention capture local nonlinear structure of distributed data rich. The robustness of the model about global fitting of data distribution, overall attention capture data, global distribution structure. The adaptability of knowledge sharing model about extraction field, explore the attention model of the unknown. A lot of research in On the basis of previous work, this paper use the inner link of the three direction between the proposed classification and new learning model, to achieve effective data mining model structure, robust learning and joint optimization. This paper has made the research results such as: first, this paper puts forward a structure Bregman distance measure function ranking algorithm.Bregman distance function based on family is a kind of flexible, nonlinear generalization of the form of the distance function. The algorithm uses the function nonlinear modeling ability of the distance learning to adapt to the Bregman distance function data structure and semantic data mining effectively distributed calibration. On the other hand, the algorithm structure the thought of learning models based on the sequence structure information sorted list to learn to adapt to the sorting task model. The algorithm provides a combined distance function A general framework for learning and learning to rank number, through the whole structure while modeling data nonlinear distribution mode and the output list, realize the joint optimization of the scheduling model and the task to adapt to the data. Two, proposed a unified taxonomy effective robust learning framework, since the step (Self-Paced Boost Learning) to enhance the learning the framework reveals and use effective learning methods and model selection to enhance robust sample selection step self learning method based on consistency and complementarity, the classification model of the formal process for discriminative model selection and robustness of sample selection. At the same time combined model by learning from weak to strong classifier and from easy to difficult learning samples can also distinguish between class model to capture is to ensure the reliability of the learning samples, the classification of learning effectiveness and robustness of the joint Enhanced. Three, proposed by the semantic relevance constraints can be adapted to the zero sample classification model of zero sample from the perspective of learning knowledge transfer problems (from the visible category learning training data on the training data of the target category classifier). Introducing semantic correlation regularization of novel Correlation (Semantic Regularization, SCR) method, meet the semantic relationship between categories through the classifier output constraints, to excavate the training categories and target class common features and semantic pattern, enhance the adaptability model of target class data. By optimizing the model sample of SCR regularization and self control joint selection based on a formalization of the discriminant the combined classifier, enhance the robustness and adaptability in cross semantic, zero sample learning achieved effective knowledge transfer.

【學(xué)位授予單位】：浙江大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2017
【分類號】：TP181

【相似文獻(xiàn)】

相關(guān)期刊論文 前8條

1 馮艷艷;;淺議圖書館知識遷移共享戰(zhàn)略[J];科技視界;2013年32期

2 李玉萍;知識遷移與相似信息共振[J];晉陽學(xué)刊;2001年01期

3 王金根,張智國;關(guān)注知識、情感的關(guān)聯(lián)和遷移[J];現(xiàn)代教學(xué);2005年Z2期

4 劉博;雷汝海;;異質(zhì)Agent間的知識遷移強(qiáng)化學(xué)習(xí)[J];中國科技論文在線;2010年02期

5 陳紅;;e-Learning培訓(xùn)的效能及其適用條件[J];教育教學(xué)論壇;2010年14期

6 林克明;薛永生;文娟;;一種基于信息論的歸納分類學(xué)習(xí)算法[J];廈門大學(xué)學(xué)報(自然科學(xué)版);2009年02期

7 王莉莉;張少白;;一種基于聲音分類學(xué)習(xí)的神經(jīng)模型研究[J];計算機(jī)技術(shù)與發(fā)展;2011年12期

8 潘巍;馬培軍;蘇小紅;;大間隔分類學(xué)習(xí)研究現(xiàn)狀[J];智能計算機(jī)與應(yīng)用;2013年06期

相關(guān)會議論文 前3條

1 李海鷹;;淺談數(shù)學(xué)在物理中的應(yīng)用[A];萃英集——青海省教育委員會、青海省教育學(xué)會優(yōu)秀教育論文集[C];2000年

2 密維才;;談?wù)Z文知識遷移的結(jié)構(gòu)層次與學(xué)法指導(dǎo)[A];萃英集——青海省教育委員會、青海省教育學(xué)會優(yōu)秀教育論文集[C];2000年

3 楊波;秦鋒;程澤凱;;一種新的分類學(xué)習(xí)系統(tǒng)評估度量[A];2005年“數(shù)字安徽”博士科技論壇論文集[C];2005年

相關(guān)重要報紙文章 前3條

1 明德實(shí)驗(yàn)學(xué)校 馮珍;提煉科學(xué)方法,優(yōu)化物理教學(xué)[N];學(xué)知報;2011年

2 王偉;新課程下的預(yù)習(xí)策略是什么？[N];中國教育報;2003年

3 記者 崔維利 袁松年;四平“分類學(xué)習(xí)法”實(shí)現(xiàn)全覆蓋[N];吉林日報;2013年

相關(guān)博士學(xué)位論文 前1條

1 皮特;基于度量學(xué)習(xí)和知識遷移的魯棒分類和排序?qū)W習(xí)研究[D];浙江大學(xué);2017年

相關(guān)碩士學(xué)位論文 前7條

1 付明杰;促進(jìn)知識遷移的專題式教學(xué)案例設(shè)計研究[D];云南大學(xué);2016年

2 周東;高中化學(xué)“烴的衍生物”中知識遷移的教學(xué)策略研究[D];浙江師范大學(xué);2011年

3 姚靜嫻;影響高中生物理知識遷移的因素及對策研究[D];河北大學(xué);2012年

4 吳玲;中學(xué)化學(xué)知識遷移的教學(xué)策略與實(shí)踐研究[D];華中師范大學(xué);2008年

5 涂莉;化學(xué)知識遷移的教學(xué)策略[D];華中師范大學(xué);2006年

6 葛亞波;分類學(xué)習(xí)中分散效應(yīng)及其認(rèn)知機(jī)制的研究[D];浙江師范大學(xué);2016年

7 張迪;基于跨領(lǐng)域分類學(xué)習(xí)的產(chǎn)品評論情感分析[D];上海交通大學(xué);2010年

，

本文編號：1770766