Publications

Detailed Information

Fast Approximate Algorithms for k-NN Search and k-NN Graph Construction : k-NN 검색 및 k-NN 그래프 생성을 위한 고속 근사 알고리즘

Cited 0 time in Web of Science Cited 0 time in Scopus
Authors

Youngki Park

Advisor
이상구
Major
공과대학 전기·컴퓨터공학부
Issue Date
2015-02
Publisher
서울대학교 대학원
Keywords
k-nearest neighbor searchk-nearest neighbor graph constructioncollaborative filteringlocality-sensitive hashing
Description
학위논문 (박사)-- 서울대학교 대학원 : 전기·컴퓨터공학부, 2015. 2. 이상구.
Abstract
Finding k-nearest neighbors (k-NN) is an essential part of recommeder systems, information retrieval, and many data mining and machine learning algorithms. However, there are two main problems in finding k-nearest neighbors: 1) Existing approaches require a huge amount of time when the number of objects or dimensions is scale up. 2) The k-NN computation methods do not show the consistent performance over different search tasks and types of data. In this dissertation, we present fast and versatile algorithms for finding k-nearest neighbors in order to cope with these problems. The main contributions are summarized as follows: first, we present an efficient and scalable algorithm for finding an approximate k-NN graph by filtering node pairs whose large value dimensions do not match at all. Second, a fast collaborative filtering algorithm that utilizes k-NN graph is presented. The main idea of this approach is to reverse the process of finding k-nearest neighbors in item-based collaborative filtering. Last, we propose a fast approximate algorithm for k-NN search by selecting query-specific signatures from a signature pool to pick high-quality k-NN candidates.The experimental results show that the proposed algorithms guarantee a high level of accuracy while also being much faster than the other algorithms over different types of search tasks and datasets.
Language
English
URI
https://hdl.handle.net/10371/119070
Files in This Item:
Appears in Collections:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share