主動(dòng)學(xué)習(xí)停止準(zhǔn)則與評(píng)價(jià)測度研究
發(fā)布時(shí)間：2018-09-11 19:43
【摘要】：主動(dòng)學(xué)習(xí)是機(jī)器學(xué)習(xí)領(lǐng)域中最為活躍的研究方向之一,其旨在花費(fèi)盡可能少的人類標(biāo)注代價(jià)獲得性能較高的分類模型。因此,在主動(dòng)學(xué)習(xí)過程中,能否定義一個(gè)合適的停止準(zhǔn)則對(duì)主動(dòng)學(xué)習(xí)是否能發(fā)揮出最大效應(yīng)具有重大意義。此外,在對(duì)一種主動(dòng)學(xué)習(xí)算法的性能進(jìn)行評(píng)估時(shí),往往需要定義一些定量的評(píng)價(jià)測度,而這正是前人工作所忽略的問題。故本文主要針對(duì)上述兩類問題展開研究。本文首先介紹了幾種常用的主動(dòng)學(xué)習(xí)停止準(zhǔn)則,進(jìn)而針對(duì)現(xiàn)有的選擇精度主動(dòng)學(xué)習(xí)停止準(zhǔn)則僅適用于批量樣例標(biāo)注場景這一缺點(diǎn),提出了一種改進(jìn)的適用于單輪單樣例標(biāo)注場景的選擇精度停止準(zhǔn)則。該準(zhǔn)則通過監(jiān)督自本輪起前溯的固定學(xué)習(xí)輪次內(nèi)的預(yù)測標(biāo)記與真實(shí)標(biāo)記間的匹配關(guān)系,對(duì)選擇精度進(jìn)行近似的評(píng)估計(jì)算,匹配度越高則選擇精度越高。繼而利用滑動(dòng)時(shí)間窗實(shí)時(shí)監(jiān)測該選擇精度的變化,若當(dāng)其高于事先設(shè)定的閾值時(shí),則停止主動(dòng)學(xué)習(xí)算法的運(yùn)行。以基于支持向量機(jī)的主動(dòng)學(xué)習(xí)方法為例,通過6個(gè)基準(zhǔn)數(shù)據(jù)集對(duì)該準(zhǔn)則的有效性與可行性進(jìn)行了驗(yàn)證,結(jié)果表明當(dāng)選取合適的閾值時(shí),該準(zhǔn)則能找到主動(dòng)學(xué)習(xí)停止的合理時(shí)機(jī)。該方法擴(kuò)大了選擇精度停止準(zhǔn)則的適用范圍,提升了其實(shí)用性。目前,適用于主動(dòng)學(xué)習(xí)的算法多種多樣,但這些主動(dòng)學(xué)習(xí)算法都共用一個(gè)統(tǒng)一的性能評(píng)估測度,即學(xué)習(xí)曲線。學(xué)習(xí)曲線在整個(gè)主動(dòng)學(xué)習(xí)迭代過程中能夠很好的區(qū)分分類模型間的性能差異,因此大多數(shù)文章都使用學(xué)習(xí)曲線作為比較不同分類算法性能的標(biāo)準(zhǔn)。但是對(duì)于兩個(gè)分類性能相近的主動(dòng)學(xué)習(xí)算法,很難從學(xué)習(xí)曲線的分布上觀察到性能變化的細(xì)微差異。針對(duì)這一問題,通過深入挖掘?qū)W習(xí)曲線中所隱藏的信息,提出了四種定量的主動(dòng)學(xué)習(xí)性能評(píng)估測度,分別為學(xué)習(xí)曲線下的面積、對(duì)數(shù)化的學(xué)習(xí)曲線下的面積、平均梯度角以及對(duì)數(shù)化的平均梯度角。在比較基于同質(zhì)分類器的主動(dòng)學(xué)習(xí)算法時(shí),這四種度量測度均能夠保證評(píng)估結(jié)果的公正性;而對(duì)于異質(zhì)的分類器,在比較不同的主動(dòng)學(xué)習(xí)算法性能時(shí),平均梯度角以及對(duì)數(shù)化的平均梯度角比另外兩種評(píng)估測度可能更加適用。此外,對(duì)數(shù)化的學(xué)習(xí)曲線下的面積與對(duì)數(shù)化的平均梯度角則更關(guān)注于主動(dòng)學(xué)習(xí)初始學(xué)習(xí)階段的性能提升速率。通過在9個(gè)數(shù)據(jù)集以及多個(gè)基準(zhǔn)主動(dòng)學(xué)習(xí)算法上的大量實(shí)驗(yàn)驗(yàn)證了上述四種測度的實(shí)用性。
[Abstract]:Active learning is one of the most active research fields in the field of machine learning. Therefore, it is of great significance to define an appropriate stop criterion in the process of active learning. In addition, when evaluating the performance of an active learning algorithm, it is often necessary to define some quantitative evaluation measures, which is the problem neglected by the previous work. Therefore, this paper mainly focuses on the above two kinds of problems. In this paper, we first introduce several commonly used active learning stopping criteria, and then aim at the disadvantage that the existing active learning stopping criteria with selective precision are only suitable for batch sample tagging scenarios. In this paper, an improved precision stopping criterion for single-wheel single-sample scene selection is proposed. By monitoring the matching relationship between prediction marks and real markers in a fixed learning cycle from the beginning of this round, the criterion evaluates and calculates the selection accuracy approximately, and the higher the matching degree is, the higher the selection accuracy is. Then the sliding time window is used to monitor the change of the selection accuracy in real time, and when the threshold is higher than the pre-set threshold, the active learning algorithm is stopped. Taking the active learning method based on support vector machine as an example, the validity and feasibility of the criterion are verified by six datum data sets. The results show that the criterion can find a reasonable time to stop active learning when the appropriate threshold is selected. This method expands the scope of application of the selective precision stop criterion and improves its practicability. At present, there are a variety of algorithms for active learning, but these active learning algorithms all share a unified performance evaluation measure, that is, learning curve. The learning curve can distinguish the performance difference between classification models well in the whole active learning iterative process, so most articles use learning curve as the standard to compare the performance of different classification algorithms. However, for two active learning algorithms with similar classification performance, it is difficult to observe the subtle variation of performance in the distribution of learning curves. In order to solve this problem, by digging the hidden information in the learning curve, four kinds of quantitative active learning performance evaluation measures are proposed, which are the area under the learning curve and the area under the logarithmic learning curve. The average gradient angle and the logarithmic average gradient angle. When comparing active learning algorithms based on homogeneous classifiers, these four metrics can ensure the fairness of the evaluation results, while for heterogeneous classifiers, when comparing the performance of different active learning algorithms, The average gradient angle and the logarithmic average gradient angle may be more suitable than the other two evaluation measures. In addition, the area under logarithmic learning curve and the average gradient angle of logarithmic learning pay more attention to the performance improvement rate in the initial learning stage of active learning. The practicability of the four measures is verified by a large number of experiments on 9 datasets and multiple benchmark active learning algorithms.
【學(xué)位授予單位】：江蘇科技大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2016
【分類號(hào)】：TP181
本文編號(hào)：2237684
[Abstract]:Active learning is one of the most active research fields in the field of machine learning. Therefore, it is of great significance to define an appropriate stop criterion in the process of active learning. In addition, when evaluating the performance of an active learning algorithm, it is often necessary to define some quantitative evaluation measures, which is the problem neglected by the previous work. Therefore, this paper mainly focuses on the above two kinds of problems. In this paper, we first introduce several commonly used active learning stopping criteria, and then aim at the disadvantage that the existing active learning stopping criteria with selective precision are only suitable for batch sample tagging scenarios. In this paper, an improved precision stopping criterion for single-wheel single-sample scene selection is proposed. By monitoring the matching relationship between prediction marks and real markers in a fixed learning cycle from the beginning of this round, the criterion evaluates and calculates the selection accuracy approximately, and the higher the matching degree is, the higher the selection accuracy is. Then the sliding time window is used to monitor the change of the selection accuracy in real time, and when the threshold is higher than the pre-set threshold, the active learning algorithm is stopped. Taking the active learning method based on support vector machine as an example, the validity and feasibility of the criterion are verified by six datum data sets. The results show that the criterion can find a reasonable time to stop active learning when the appropriate threshold is selected. This method expands the scope of application of the selective precision stop criterion and improves its practicability. At present, there are a variety of algorithms for active learning, but these active learning algorithms all share a unified performance evaluation measure, that is, learning curve. The learning curve can distinguish the performance difference between classification models well in the whole active learning iterative process, so most articles use learning curve as the standard to compare the performance of different classification algorithms. However, for two active learning algorithms with similar classification performance, it is difficult to observe the subtle variation of performance in the distribution of learning curves. In order to solve this problem, by digging the hidden information in the learning curve, four kinds of quantitative active learning performance evaluation measures are proposed, which are the area under the learning curve and the area under the logarithmic learning curve. The average gradient angle and the logarithmic average gradient angle. When comparing active learning algorithms based on homogeneous classifiers, these four metrics can ensure the fairness of the evaluation results, while for heterogeneous classifiers, when comparing the performance of different active learning algorithms, The average gradient angle and the logarithmic average gradient angle may be more suitable than the other two evaluation measures. In addition, the area under logarithmic learning curve and the average gradient angle of logarithmic learning pay more attention to the performance improvement rate in the initial learning stage of active learning. The practicability of the four measures is verified by a large number of experiments on 9 datasets and multiple benchmark active learning algorithms.
【學(xué)位授予單位】：江蘇科技大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2016
【分類號(hào)】：TP181
【相似文獻(xiàn)】
相關(guān)期刊論文 前7條
1 左相國,王曉明;學(xué)習(xí)曲線在工程技術(shù)改造中的應(yīng)用[J];科技進(jìn)步與對(duì)策;2004年09期
2 ;蘋果iPhone的學(xué)習(xí)曲線[J];軟件世界;2007年17期
3 王歡歡;王寧;萬小蘭;;結(jié)合學(xué)習(xí)曲線分析企業(yè)信息化成本[J];科技視界;2014年01期
4 ;實(shí)施手記之十一:ERP學(xué)習(xí)曲線[J];IT經(jīng)理世界;2000年21期
5 王科;我國集成電路企業(yè)學(xué)習(xí)曲線的實(shí)證研究初探[J];科研管理;2001年03期
6 馬慶賀;孟子厚;;聽音訓(xùn)練對(duì)漢語單音節(jié)聽感清晰度的影響[J];聲學(xué)技術(shù);2014年02期
7 ;[J];;年期
相關(guān)碩士學(xué)位論文 前10條
1 吳如陳;微創(chuàng)經(jīng)椎間孔腰椎間融合術(shù)治療腰椎退變性疾病的學(xué)習(xí)曲線評(píng)估[D];福建醫(yī)科大學(xué);2015年
2 鳳鳳;腹腔鏡肝切除術(shù)學(xué)習(xí)曲線的研究[D];中國人民解放軍醫(yī)學(xué)院;2015年
3 梁俊杰;完全乳暈入路腔鏡甲狀腺切除術(shù)持鏡者的學(xué)習(xí)曲線[D];暨南大學(xué);2015年
4 朱莉;考慮學(xué)習(xí)曲線的項(xiàng)目人力資源分配研究[D];哈爾濱工業(yè)大學(xué);2015年
5 周亮;我國風(fēng)機(jī)制造業(yè)學(xué)習(xí)曲線研究[D];廣西科技大學(xué);2015年
6 張馨月;高強(qiáng)度聚焦超聲治療子宮肌瘤的學(xué)習(xí)曲線研究[D];重慶醫(yī)科大學(xué);2016年
7 魯超;腹腔鏡胰十二指腸切除術(shù)的學(xué)習(xí)曲線[D];浙江大學(xué);2016年
8 楊菊;主動(dòng)學(xué)習(xí)停止準(zhǔn)則與評(píng)價(jià)測度研究[D];江蘇科技大學(xué);2016年
9 劉曉勇;電子類上市公司學(xué)習(xí)曲線實(shí)證研究[D];長沙理工大學(xué);2007年
10 謝榕城;經(jīng)皮內(nèi)鏡下胃造瘺術(shù)的學(xué)習(xí)曲線[D];福建醫(yī)科大學(xué);2010年
，本文編號(hào)：2237684
本文鏈接：http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/2237684.html
最近更新
教材專著
