適用於多特性多用途的分散式關連分群機制;A Distributed Correlation Based Mechanism for Adaptive and Divergent Purposed Clustering

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/85060

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/85060

Title:	適用於多特性多用途的分散式關連分群機制;A Distributed Correlation Based Mechanism for Adaptive and Divergent Purposed Clustering
Authors:	李桂昇;Lee, Kuei-Sheng
Contributors:	資訊工程學系
Keywords:	大型資料;分群演算法;分散式系統;機械學習;Big Data;Clustering;Distributed system;Machine learning
Date:	2021-01-18
Issue Date:	2021-03-18 17:32:05 (UTC+8)
Publisher:	國立中央大學
Abstract:	在機械學習的領域中，分群分析（Cluster analysis）一直是很重要的一門技術。資料分群後會讓具有相似特性的單元聚類在一起，從而得知其中有用或隱含的訊息。然而目前主流的分群分析演算法皆需要全面性分析整體資料以取得演算法中的最佳參數，如此一來，面對大型資料的處理將難以施展。本研究提出一種非監督式學習 (Unsupervised Learning）分散式關連分群機制。假設同一群中相鄰資料點皆為二二相似，則可依此特性關連至更多的資料點而為一個完整的群組。而在處理資料的時候，可將大型資料拆解分散至多台電腦，平行計算任二筆資料間的關連，之後再過濾及彙整處理結果集結為群組。在本研究實作時使用了二維圖形、圍棋分析及醫學資料做為實驗數據，依資料類型不同分別訂定了相似性計算方式。實驗結果顯示出此分群機制處理大型資料的能力，同時也提供了良好的執行效能，更有其準確性、適用性及易用性等特性。 ;Cluster analysis is an important technique in the field of machine learning. Data clustering allows units with similar characteristics to be clustered together in order to learn useful or implicit information. However, current mainstream cluster analysis algorithms need to analyze the whole dataset comprehensively to obtain the best parameters, which makes it difficult to process large-scale dataset. This study proposes a distributed correlation-based clustering mechanism based on unsupervised learning. If neighboring data points in the same group are similar, then they can be related to more data points to form a complete cluster according to this characteristic. In processing the data, a large-scale dataset can be disassembled and distributed to multiple computers to calculate the correlation between any two pieces of data in parallel, and then the results are filtered and aggregated into a cluster. This study uses 2D graphics, Go game (Weiqi) analysis, and medical data as experimental data, and similarity calculations are developed according to the data types. The experimental results show the ability of this clustering mechanism to handle large-scale dataset. This clustering mechanism provides advantages such as good execution performance, accuracy, variability, applicability, and ease of use.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	121	View/Open

社群 sharing

Loading...