非均衡加權(quán)隨機(jī)梯度下降SVM在線算法
發(fā)布時間:2018-08-04 07:46
【摘要】:隨機(jī)梯度下降(stochastic gradient descent,SGD)方法已被應(yīng)用于大規(guī)模支持向量機(jī)(support vector machine,SVM)訓(xùn)練,其在訓(xùn)練時采取隨機(jī)選點的方式,對于非均衡分類問題,導(dǎo)致多數(shù)類點被抽取到的概率要遠(yuǎn)遠(yuǎn)大于少數(shù)類點,造成了計算上的不平衡。為了處理大規(guī)模非均衡數(shù)據(jù)分類問題,提出了加權(quán)隨機(jī)梯度下降的SVM在線算法,對于多數(shù)類中的樣例被賦予較小的權(quán)值,而少數(shù)類中的樣例被賦予較大的權(quán)值,然后利用加權(quán)隨機(jī)梯度下降算法對SVM原問題進(jìn)行求解,減少了超平面向少數(shù)類的偏移,較好地解決了大規(guī)模學(xué)習(xí)中非均衡數(shù)據(jù)的分類問題。
[Abstract]:Stochastic gradient descent (stochastic gradient descenting (stochastic gradient) method has been applied to large-scale support vector machine (support vector machine) training. The probability of extracting most of the points is much higher than that of a few, which results in the imbalance of calculation. In order to deal with the problem of large-scale disequilibrium data classification, a weighted stochastic gradient descent SVM online algorithm is proposed, in which the sample in most classes is given a smaller weight, while the sample in a few classes is given a larger weight. Then the weighted stochastic gradient descent algorithm is used to solve the original SVM problem, which reduces the deviation of the superplane for a few classes and solves the classification problem of unbalanced data in large-scale learning.
【作者單位】: 河北大學(xué)數(shù)學(xué)與信息科學(xué)學(xué)院河北省機(jī)器學(xué)習(xí)與計算智能重點實驗室;
【基金】:河北省自然科學(xué)基金No.F2015201185~~
【分類號】:TP181
本文編號:2163139
[Abstract]:Stochastic gradient descent (stochastic gradient descenting (stochastic gradient) method has been applied to large-scale support vector machine (support vector machine) training. The probability of extracting most of the points is much higher than that of a few, which results in the imbalance of calculation. In order to deal with the problem of large-scale disequilibrium data classification, a weighted stochastic gradient descent SVM online algorithm is proposed, in which the sample in most classes is given a smaller weight, while the sample in a few classes is given a larger weight. Then the weighted stochastic gradient descent algorithm is used to solve the original SVM problem, which reduces the deviation of the superplane for a few classes and solves the classification problem of unbalanced data in large-scale learning.
【作者單位】: 河北大學(xué)數(shù)學(xué)與信息科學(xué)學(xué)院河北省機(jī)器學(xué)習(xí)與計算智能重點實驗室;
【基金】:河北省自然科學(xué)基金No.F2015201185~~
【分類號】:TP181
【相似文獻(xiàn)】
相關(guān)期刊論文 前3條
1 謝漢中;在PLA里交叉點故障定位的在線算法[J];江蘇電機(jī)工程;1987年01期
2 帥典勛;在可編程序的邏輯陣列(PLA)中交叉點故障定位的一種在線算法[J];計算機(jī)工程;1984年06期
3 賀文武;;在線核學(xué)習(xí)的一般形式探討[J];福建工程學(xué)院學(xué)報;2010年04期
相關(guān)會議論文 前2條
1 何宇新;韓志剛;;多層遞階方法時變參數(shù)估值最佳初值和最佳跟蹤的在線算法[A];1992年中國控制與決策學(xué)術(shù)年會論文集[C];1992年
2 尹煥平;孫宗海;;基于自然梯度的支持向量回歸在線算法[A];2009中國控制與決策會議論文集(3)[C];2009年
相關(guān)博士學(xué)位論文 前1條
1 黃禹瀟;面向復(fù)雜診斷貝葉斯網(wǎng)實時推理問題的離線、在線算法的研究[D];吉林大學(xué);2012年
相關(guān)碩士學(xué)位論文 前1條
1 王明岳;m射線路徑上移動目標(biāo)搜索的在線算法研究[D];復(fù)旦大學(xué);2009年
,本文編號:2163139
本文鏈接:http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/2163139.html
最近更新
教材專著