English  |  正體中文  |  简体中文  |  Items with full text/Total items : 26988/38789
Visitors : 2311827      Online Users : 48
RC Version 4.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
Scope Adv. Search
LoginUploadHelpAboutAdminister

Please use this identifier to cite or link to this item: http://ntour.ntou.edu.tw:8080/ir/handle/987654321/49311

Title: 基於在流形結構上隨機漫步的主動式學習演算法
Authors: Liao, Bo-Han
廖柏翰
Contributors: NTOU:Department of Computer Science and Engineering
國立臺灣海洋大學:資訊工程學系
Keywords: 流形學習;主動式學習;隨機漫步;局部線性內嵌法
Manifold learning;Active learning;Random walks;Locally linear embedding
Date: 2015
Issue Date: 2018-08-22T06:56:31Z
Abstract: 這篇論文主要探討機器學習領域中的主動式學習,一個主動式學習必須要在眾多的資料當中,選取到我們認為重要並具有代表性的資料點。幾十年來,已經有很多主動式學習方法陸續被提出,若使用者對於資料集有充分的了解,挑選了適當的方法與相對應的參數後,便能在該資料集中挑選到適當的資料,達到使用者的需求。但是當一個資料集需要主動式學習時,我們通常都對資料的分布與結構一無所知,我們並沒有辦法容易地挑選到適當的參數。因此,我們提出了一個演算法可以探索資料分佈所形成的流形結構,將結構上重要資料點挑選出來讓使用者標籤。其中,參數全部由演算法自動計算並挑選。同時,這演算法也能夠處理規模數萬筆等級的資料。實驗結果已驗證本論文所提的演算法的效能與可行性。
This paper considers the active learning problem of selecting important instances from the unlabeled data pool. Recently, many active learning algorithms have been proposed to select data instances to acquire class labels for subsequent supervised or semi-supervised learning. For many state-of-the-art methods, to select query samples properly, several parameters for the kernel function and for the construction of the nearest neighbor graph, which are critical to the quality of the selected query sample, are needed. However, choosing these parameters before exploring the data set is difficult. Thus, based on an assumption that manifold structure of the data is local and sparse, we proposed an algorithm that learns the manifold structure by a kernel version of local linear embedding and can tune the kernel parameter automatically. The scalability of our proposed algorithm has been proven by testing the proposed algorithm against several large-scale data sets. Experimental results have shown that the samples selected by the proposed algorithm are informative in comparison with several state-of-the-art active learning methods.
URI: http://ethesys.lib.ntou.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=G0010257014.id
http://ntour.ntou.edu.tw:8080/ir/handle/987654321/49311
Appears in Collections:[資訊工程學系] 博碩士論文

Files in This Item:

There are no files associated with this item.



All items in NTOUR are protected by copyright, with all rights reserved.

 


著作權政策宣告: 本網站之內容為國立臺灣海洋大學所收錄之機構典藏,無償提供學術研究與公眾教育等公益性使用,請合理使用本網站之內容,以尊重著作權人之權益。
網站維護: 海大圖資處 圖書系統組
DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - Feedback