离散局部线性嵌入哈希-计算机技术专业毕业论文.docxVIP

AbstractThe Abstract The efficient indexing and searching becomes a hot research topic due to the explosive growth in big data．Locating the nearest neighbors for a query for large scale searching and pattern matching is a crucial problem in this research area．Considering the memory consumption and time cost，it is unpractical to adopt exhaustive comparison to handle this task．Due to the strong capacity(in terms of both effectiveness and efficiency)of hashing based approximate nearest neighbor(ANN)search，ANN becomes a very popular choice to deal with this problem．With hashing techniques，the original data is transformed to a new space(Hamming space)，while the neighborhood relationship Call still be maintained． In this paper,we propose a novel hashing algorithm named Discrete Locally Linear Embedding(DLLE)．We reconstruct the data points in the Hamming Space，thus the learned hashing function carl maintain the local linear relationship of the original data．We solve the problem directly in the hamming space whithout losing the precision．However,the DLLE algorithm needs to compute the local linear relationship matrix for the original data,which is time expensive．So we furtherly propose Anchor Discrete Locally Linear Embedding (ADLLE)algorithm based on DLLE．The local linear relationship matrix of the original data is represented by the product of low rank matirx which is time efficient． Key Words：Hashing，Discrete Locality Linear Embedding，Anchor III 万方数据 目 目 录 摘要．．．．．．．．．．．．·-···-····一一一一一一一一 I 第1章耋嗽．．．．··．·········一一一一一一一一·1 1．1研究背量．．．．．．．．．···········一-．一一一一 1 1．3本文主要研究工作．．．．················一．．一· 5 第2章相关工作．．．．．·．······················· 7 2．2基于哈希的ANN搜素流程．．．．·．················· 7 2．2．1设计哈希函数．．．．．．．．．．．．············一···8 2．2．2建立哈希表．．．．．．．．．·．·．·················8 2．2．3使用哈希码进行在线查询．．．．．．．．．·．············ 9 2．3．1局部敏感哈希LSH．．．．．·．·．················· 9 2．3．2 MinHash 10 2．3．3迭代量化ITQ． 11 2．3．4锚图哈希．．．．．．．．．．····················· 11 2．4本章小结．．．．．．．．．．．·················-··· 13 V 万方数据 目