UTokyo Repository 東京大学

UTokyo Repository >
122 新領域創成科学研究科 >
14 基盤科学研究系 複雑理工学専攻 >
1221420 博士論文(基盤科学研究系複雑理工学専攻) >

このページ(論文)をリンクする場合は次のURLを使用してください: http://hdl.handle.net/2261/52457

タイトル: Analysis and Applications of the T-complexity
その他のタイトル: T-complexityの解析と応用
著者: Hamano, Kenji
著者(別言語): 濱野, 健二
キーワード: T-code
Lempel-Ziv complexity
NIST SP 800-22
data compression
complexity profile
発行日: 2010年3月24日
抄録: T-codes are variable-length self-synchronizing codes introduced by Titchener in 1984. T-code codewords are constructed recursively from a finite alphabet using an algorithm called T-augmentation, resulting in excellent self-synchronization properties. An algorithm called T-decomposition parses a given sequence into a series of T-prefixes, and finds a T-code set in which the sequence is encoded to a longest codeword. There are similarities and differences between T-decomposition and the conventional LZ78 incremental parsing. The LZ78 incremental parsing algorithm parses a given sequence into consecutive distinct subsequences (words) sequentially in such a way that each word consists of the longest matching word parsed previously and a literal symbol. Then, the LZ-complexity is defined as the number of words. By contrast, T-decomposition parses a given sequence into a series of T-prefixes, each of which consists of the recursive concatenation of the longest matching T-prefix parsed previously and a literal symbol, and it has to access the whole sequence every time it determines a T-prefix. Alike to the LZ-complexity, the T-complexity of a sequence is defined as the number of T-prefixes, however, the T-complexity of a particular sequence in general tends to be smaller than the LZ-complexity. In the first part of the thesis, we deal with our contributions to the theory of T-codes. In order to realize a sequential determination of T-prefixes, we devise a new T-decomposition algorithm using forward parsing. Both the T-complexity profile obtained from the forward T-decomposition and the LZ-complexity profile can be derived in a unified way using a differential equation method. The method focuses on the increase of the average codeword length of a code tree. The obtained formulas are confirmed to coincide with those of previous studies. The magnitude of the T-complexity of a given sequence s in general indicates the degree of randomness. However, there exist interesting sequences that have much larger T-complexities than any random sequences. We investigate the maximum T-complexity sequences and the maximum LZ-complexity sequences using various techniques including those of the test suite released by the National Institute of Standards and Technology (NIST) of the U.S. government, and find that the maximum T-complexity sequences are less random than the maximum LZ-complexity sequences. In the second part of the thesis, we present our achievements in terms of application. We consider two applications -- data compression and randomness testing. First, we propose a new data compression scheme based on T-codes using a dictionary method such that all phrases added to a dictionary have a recursive structure similar to T-codes. Our data compression scheme can compress any of the files in the Calgary Corpus more efficiently than previous schemes based on T-codes and the UNIX compress, a variant of LZ78 (LZW). Next, we introduce a randomness test based on the T-complexity. Recently, the Lempel-Ziv (LZ) randomness test based on the LZ-complexity was officially excluded from the NIST test suite. This is because the distribution of P-values for random sequences of length 106, the most common length used, is strictly discrete in the case of the LZ-complexity. Our test solves this problem because the T-complexity features an almost ideal uniform continuous distribution of P-values for random sequences of length 106. The proposed test outperforms the NIST LZ test, a modified LZ test proposed by Doganaksoy and Göloglu, and all other tests included in the NIST test suite, in terms of the detection of undesirable pseudorandom sequences generated by a multiplicative congruential generator (MCG) and non-random byte sequences Y = Y0, Y1, Y2, ···, where Y3i and Y(3i+1) are random, but Y(3i+2) is given by Y3i + Y(3i+1) mod 28.
内容記述: 報告番号: 甲26141 ; 学位授与年月日: 2010-03-24 ; 学位の種別: 課程博士 ; 学位の種類: 博士(科学) ; 学位記番号: 博創域第558号 ; 研究科・専攻: 新領域創成科学研究科基盤科学研究系複雑理工学専攻
URI: http://hdl.handle.net/2261/52457
出現カテゴリ:021 博士論文
1221420 博士論文(基盤科学研究系複雑理工学専攻)


ファイル 記述 サイズフォーマット
K-02429.pdf本文(fulltext)2.95 MBAdobe PDF見る/開く



Valid XHTML 1.0! DSpace Software Copyright © 2002-2010  Duraspace - ご意見をお寄せください