Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/37483
Title: | 實作以綑集法解線性支持向量機問題 An Implementation of the Bundle Method |
Authors: | Mi-Chen Tsai 蔡宓真 |
Advisor: | 林智仁(Chih-Jen Lin) |
Keyword: | 線性支持向量機,綑集法,切面法,副梯度,大規模稀疏資料分類, linear support vector machines,bundle method,cutting plane,subgradient,large-scale sparse data classification, |
Publication Year : | 2008 |
Degree: | 碩士 |
Abstract: | 稀疏資料分類的問題近年來在文件分類與自然語言處理等領域中很常見,線性支持向量機對於大規模稀疏資料分類便漸趨實用。傳統以斜率為基礎的方法無法用於解決一維範數損失函數支持向量機(L1-SVM)的問題,於是諸如綑集法和切面的技巧等就被利用在這類不可微分的問題上。在這篇論文中,我們在這篇論文中利用 libsvm 實作了 Smola et al. (2008) 中提出的綑集法,我們也列出了一些實驗上與 bmrm 函式庫的比較。 Classification on data with sparse features are common in document classification and natural language processing. Linear support vector machines (SVM) thus is for classifying large-scale sparse data. Some optimization formulations like L1-SVM cannot be minimized with traditional gradient based approaches. Methods like bundle methods and cutting plane techniques are useful for such non-differentiable SVM problems. In this thesis, we implement the bundle method proposed in Smola et al. (2008) by modifying libsvm. We also experimentally compare our implementation with another implementation mrm. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/37483 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 資訊工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-97-1.pdf Restricted Access | 1.04 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.