Advance

SHIRATSUCHI Hiroshi

    Department of Electric and Electronic Engineering　Professor
Last Updated :2024/04/19

Researcher Information

Degree

  • (BLANK)

J-Global ID

Research Interests

  • ニューラルネットワーク   信号処理   Neural Network   

Research Areas

  • Manufacturing technology (mechanical, electrical/electronic, chemical engineering) / Communication and network engineering
  • Informatics / Intelligent informatics

Academic & Professional Experience

  • 2024/04 - Today  Kindai UniversityFaculty of Humanity-Oriented Science and Engineering Department of Electric and Electronic EngineeringProfessor
  • 2016/04 - 2024/03  Kindai UniversityFaculty of Humanity-Oriented Science and Engineering Department of Electric and Electronic Engineering准教授
  • 2009/04 - 2016/03  Kindai UniversityFaculty of Humanity-Oriented Science and Engineering准教授
  • 2005/04 - 2009/03  Kindai UniversityFaculty of Humanity-Oriented Science and Engineering講師
  • 1999/04 - 2005/03  University of the RyukyusFaculty of Engineering, Department of Information Engineering助手

Education

  • 1996/04 - 1999/03  Kyushu Institute of Technology  情報工学研究科  情報科学専攻

Association Memberships

  • 電子情報通信学会   

Published Papers

  • Matsuzaki Takanori; Taguchi Emiko; Horiuchi Kozo; Shiratsuchi Hiroshi; Mashiko Koichiro
    ICIC Express Letters, Part B: Applications ICIC International 13 (06) 589  2185-2766 2022
  • Shiratsuchi Hiroshi; Furuta Misato; Wakamori Yuji; Horiuchi Kozo; Matsuzaki Takanori
    ICIC Express Letters ICIC International 16 (06) 621  1881-803X 2022
  • Development of Web-Based Arduino Simulator Supporting Debug Function and Status Visualization
    Kozo Horiuchi; Takanori Matsuzaki; Hiroshi Shiratsuchi
    ICIC Express Letters, Part B: Applications 11 (9) 815 - 822 2020/09 [Refereed]
  • Proposal of Soft Error Injection Method Using Two FPGA Boards
    Takanori Matsuzaki; Ken Miyashita; Kozo Horiuchi; Hiroshi Shiratsuchi; Ili Shairah; Abdul Halim; Koichiro Mashiko
    ICIC Express Letters 14 (6) 555 - 562 2020/06 [Refereed]
  • Studies on Development of Web-Based Integrated Learning and Education Support System
    Hiroshi Shiratsuchi; Kozo Horiuchi; Takanori Matsuzaki
    ICIC Express Letters, Part B: Applications 11 (2) 197 - 206 2020/02 [Refereed]
  • MATSUZAKI Takanori; Osman Elfadil; HORIUCHI Kozo; SHIRATSUCHI Hiroshi; MASHIKO Koichiro
    International Journal of Innovative Computing, Information and Control 14 (3) 997 - 1008 2019/03 [Refereed]
  • A Study on Usage Status Survey and User Location History of Campus Wireless LAN System
    H.Shiratsuchi; K.Horiuchi; T.Matsuzaki
    ICIC Express Letters, Part B: Applications 10 (2) 121 - 128 2019/02 [Refereed]
  • SHIRATSUCHI Hiroshi; HORIUCHI Kozo; MATSUZAKI Takanori
    ICIC Express Letters, Part B:Applications 10 (2) 121 - 127 2019/02 [Refereed]
  • HORIUCHI Kozo; MATSUZAKI Takanori; SHIRATSUCHI Hiroshi
    ICIC Express Letters, Part B:Applications 10 (2) 137 - 146 2019/02 [Refereed]
  • SHIRATSUCHI Hiroshi; HORIUCHI KOZO
    Journal of Japan Society for Engineering Education 66 (5) 37 - 43 2018/10 [Refereed]
  • SHIRATSUCHI Hiroshi; HAMANA Yukiya; HORIUCHI Kozo; KYURA Nobuhiro
    ICIC Express Letters, Part B:Applications 8 (12) 1585 - 1594 2017/12 [Refereed]
  • SOGO Hayato; TANAKA Hironori; IWASAKI Nobuo; MATSUZAKI Takanori; SHIRATSUCHI Hiroshi; GOTANDA Hiromu
    ICIC Express Letters, Part B:Applications ICIC International 6 (3) 627 - 635 2185-2766 2015/03 [Refereed]
  • Hiroshi Shiratsuchi; Hiromu Gotanda
    2015 9TH INTERNATIONAL CONFERENCE ON SIGNAL PROCESSING AND COMMUNICATION SYSTEMS (ICSPCS) IEEE 1 - 8 2015 [Refereed]
     
    It is well known that, in the case of frequency domain blind channel estimation, the channel gains are correctly estimated but the channel phases are not necessarily estimated correctly: the channel estimates are often accompanied with phase ambiguity because the phase range for numerical calculation is restricted to (-pi, pi]. In this paper, a new frequency-domain blind channel estimation without phase ambiguity is proposed for QAM-OFDM systems of which transmitting and receiving stations are in LoS (Line of Sight) under the conditions: i) the absolute value of phase rotation is less than pi/2 at all subcarriers, and ii) it is less than pi/4 for at least one subcarrier among them. From several simulation results for 50 sub-carriers with 16 QAM using 2048 OFDM blocks, it is confirmed that, when the SNR (Signal-to-Noise Ratio) in baseband region is over 20[dB], the QAM-OFDM channel is estimated at high precision without phase ambiguity. The restored symbols resulting from the channel estimates give the bit-error-rate of about 10(-3) or less at SNR >= 20[dB] for the case that the ratio of the root-mean-square delay spread to the OFDM symbol length is 1/16.
  • IWASAKI Nobuo; MATSUZAKI Takanori; Hirano Go; SHIRATSUCHI Hiroshi; INOUE Katsuhiro; GOTANDA Hiromu
    ICIC Express Letters, Part B:Applications ICIC International 5 (2) 377 - 386 2185-2766 2014/04 [Refereed]
  • TANAKA Hironori; SOGO Hayato; IWASAKI Nobuo; MATSUZAKI Takanori; SHIRATSUCHI Hiroshi; GOTANDA Hiromu
    Journal of Signal Process 18 (2) 77 - 88 1880-1013 2014 [Refereed]
  • Semi-Blind Channel Estimation for QAM-OFDM System
    五反田 博; 松崎 隆哲; 白土 浩; 十川勇人; 田中 宏典
    The 45th ISCIE International Symposium on Stochastic Systems Theory and Its Applications SB2-4 2013/11 [Refereed]
  • ステレオマイクによる音源方向のリアルタイム推定
    岩崎宣生; 井上勝裕; 石橋孝昭; 松崎隆哲; 平野剛; 白土浩; 五反田博
    2011年度日本応用数理学会研究部会連合発表会 2012/03
  • 白土 浩; 五反田 博
    Kayanomori 近畿大学産業理工学部 (14) 28 - 36 1349-5801 2011/07 
    [Abstract] Orthogonal frequency division multiplex (OFDM) systems are widely used in various fields such as wireless LAN, digital terrestrial television and 3.9th generation cellular phone due to their excellent characteristics; high frequency efficiency, high tolerance to multi-path fading and high communication stability. However, it is not easy to understand the principle of OFDM. In this paper, an OFDM simulator available for beginners is developed by means of Simulink so that they could understand the principle.
  • OFDMにおけるキャリア間干渉のICAによる解消について
    五反田 博; 白土 浩; 原谷 直実
    電子情報通信学会技術研究報告 SIP07 23 - 28 2010
  • Takeshi Koya; Nobuo Iwasaki; Takaaki Ishibashi; Go Hirano; Hiroshi Shiratsuchi; Hiromu Gotanda
    Journal of Advanced Computational Intelligence and Intelligent Informatics Fuji Technology Press 14 (4) 364 - 374 1883-8014 2010 [Refereed]
     
    In real world environments where acoustic signals are contaminated with various noises, it is difficult to estimate the Signal-to-Noise Ratio (SNR) only from signals observed at microphones the knowledge of acoustic transfer functions and original source signals is inevitable for SNR estimation. The present paper proposes a method to estimate SNR approximately in the real world environments without the knowledge of transfer functions and source signals: SNR is estimated after application of Independent Component Analysis (ICA) to the signals observed at microphones. Our proposed method also works as a speech segment detector since detection of speech segments are necessarily carried out in the course of SNR estimation. From several experimental results, the proposed method has been confirmed to be valid.
  • Hiroshi Shiratsuchi; Nobuo Iwasaki; Hironori Tanaka; Takaaki Ishibashi; Naomi Haratani; Yosimasa Nakano; Hiromu Gotanda
    TENCON 2010: 2010 IEEE REGION 10 CONFERENCE IEEE 1330 - 1335 2159-3442 2010 [Refereed]
     
    In this paper we propose a blind estimation of fading coefficients without using the pilot symbols in QAM-OFDM systems subjected to carrier frequency offset (CFO), where the phases of the fading coefficients are within +/-pi/4. First, under a frequency selective fading channel, the inter-carrier interference (ICI) due to CFO is formulated in the framework of independent component analysis (ICA). Then the permutation and scale uncertainty inherent in ICA is resolved and CFO is estimated. Next, based on the CFO estimation results, the fading coefficients are estimated by use of the fact that symbol constellation at each sub-carrier is stretched/shrunk and rotated according to the value of its fading coefficient. Then, the unknown transmitted symbols are restored by use of the coefficient estimates. Finally, the validity of the proposed approach is confirmed from several simulations for 16QAM-OFDM.
  • KOYA Takeshi; ISHIBASHI Takaaki; SHIRATSUCHI Hiroshi; GOTANDA Hiromu
    Transactions of the Institute of Systems, Control and Information Engineers THE INSTITUTE OF SYSTEMS, CONTROL AND INFORMATION ENGINEERS (ISCIE) 22 (8) 287 - 294 1342-5668 2009/08 
    In order to solve the blind source deconvolution, a convolved mixing process in the time domain is often transformed into an instantaneous mixing model in the frequency domain. However, the model is only an approximated one and thus does not work effectively under highly reverberant environments. By dividing the impulse response properly, Servi?re has precisely transformed the time-domain convolved mixture to a frequency-domain convolved mixture and has proposed a new FDICA approach available under high reverberation. In the approach, however, the permuation and scaling problems are unresol...
  • SAYVISITH Vithaya; KIMURA Tetsuya; NAKAGAWA Fuminari; SHIRATSUCHI Hiroshi; HARATANI Naomi; GOTANDA Hiromu
    The Transactions of the Institute of Electronics, Information and Communication Engineers. A The Institute of Electronics, Information and Communication Engineers 92 (3) 141 - 149 0913-5707 2009/03 
    本論文では,周波数オフセット(CFO : Carrier Frequency Offset)と周波数選択性フェージングの影響を受けたQAM-OFDMについて,フェージング係数の位相が±π/4未満のとき,フェージング係数をパイロットシンボルを使わずにブラインド的に推定する方法を提案する.具体的には,まず,周波数選択性フェージングのあるもとでCFOに起因するキャリヤ間干渉(ICI : Inter-Carrier Interference)をICA(lndependent Component Analysis)モデルで定式化して,ICAに付随する成分置換やスケール不定性の解消法やCFOの推定法を提案する.次に,各サブキャリヤのコンステレーションがフェージング係数の値に応じて伸縮・回転することを利用して,CFOの推定結果をもとにフェージング係数を推定する.そして,その係数の推定結果を用いて未知の送信シンボルを復元する.最後に,提案法の有効性を16QAM-OFDMに対するシミュレーションで確認する.
  • SAYVISITH Vithaya; KIMURA Tetsuya; NAKAGAWA Fuminari; SHIRATSUCHI Hiroshi; HARATANI Naomi; GOTANDA Hiromu
    Proc. the 40th ISCIE International Symposium on Stochastic Systems Theory and Its Applications 204 - 209 2008/11 [Refereed]
  • NAKAGAWA Fuminari; TAKASE Shigefumi; SHIRATSUCHI Hiroshi; GOTANDA Hiromu
    The Transactions of the Institute of Electronics, Information and Communication Engineers. A The Institute of Electronics, Information and Communication Engineers 91 (4) 448 - 457 0913-5707 2008/04 [Refereed]
     
    直交周波数分割多重は,周波数オフセット(CFO)がある場合,サブキャリヤ間の直交性が崩れるため,サブキャリヤ間干渉(ICI)を引き起こして通信性能が劣化する.そのため,様々な周波数オフセット推定法が開発されている.本論文では,新しく独立成分分析(ICA)に基づいてICIを解消する方法を提案する.すなわち,まず,ICI問題をICAモデルの枠組みで定式化する.次に,ICAに付随して起きる成分置換やスケール不定の解法原理を明確にして,拘束条件付き自然こう配(NG)アルゴリズムを提案する.最後に,ICAによる結果を利用して,CFOを推定するとともに,未知の送信シンボルを復元する.提案法の有効性はシミュレーションにより確認した.
  • SHIRATSUCHI Hiroshi; GOTANDA Hiromu; INOUE Katsuhiro; KUMAMARU Kousuke
    Journal of Advanced Computational Intelligence and Intelligent Informatics 12 (1) 57 - 62 2008/01 [Refereed]
     
    本論文では階層型ニューラルネットワークの学習によって得られる最終的なネットワーク構造から入出力間の関連性をルールとして抽出するための学習である構造学習において最適構造が得られやすくなるための初期条件について考察し，初期値設定法として提案した
  • 白土 浩; 五反田 博
    近畿大学産業理工学部研究報告(かやのもり) 近畿大学産業理工学部 (3) 15 - 20 1349-5801 2005/04 
    本稿て？は階層型ニューラルネットワークにおける冗長ユニットをユニット間の類似度に基つ？いた削除アルコ？
    リス？ムを提案し,アヤメ分類問題を対象としたシミュレーションにより有効性を検証し，初期ユニット数か？増加しても最終的なネッ ト構造か？変化しないことを示した．
  • GOTANDA Hiromu; UEDA Yoshihiro; SHIRATSUCHI Hiroshi
    Journal of Advanced Computational Intelligence and Intelligent Informatics 8 (6) 627 - 632 2004/12 [Refereed]
     
    階層型ニューラルネットワークを対象とした構造付き忘却学習において著者らの提案する初期値設定法を適用した際の構造決定過程の解析と生成されるネットワーク構造の最適性について示した．
  • GOTANDA Hiromu; UEDA Yoshihiro; SHIRATSUCHI Hiroshi
    Journal of Advanced Computational Intelligence and Intelligent Informatics 8 (6) 621 - 626 2004/12 [Refereed]
     
    階層型ニューラルネットワークの学習特性を解析し，これまで実験的に示されてきた両極型ユニットによる学習性能の向上についてニューラルネットワークの各ユニットが成す分離超平面による幾何学的な解析により理論的に明らかにした．
  • SHIRATSUCHI Hiroshi; GOTANDA Hiromu; INOUE Katsuhiro; KUMAMARU Kousuke
    Transactions of the Institute of Systems,Control and Information Engineers THE INSTITUTE OF SYSTEMS, CONTROL AND INFORMATION ENGINEERS (ISCIE) 15 (10) 532 - 538 1342-5668 2002/10 [Refereed]
     
    本論文では三層のニューラルネットを対象に幾何学的観点に基づいた初期パラメータの設定指針を提案すると同時に訓練データの正規化について考察し，正規化と初期値設定の関連性を明らかにした．さらにアフィン変換による正規化を学習終了後のネット荷重に埋め込む手法も提案した．
  • H Shiratsuchi; H Gotanda; K Inoue; K Kumamaru
    IJCNN'01: INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS, VOLS 1-4, PROCEEDINGS IEEE 2644 - 2649 1098-7576 2001 [Refereed]
     
    In this paper, we propose an initialization method of multilayer neural networks(NN) employing the structural learning with forgetting. The proposed initialization consists of two steps: weights of hidden units are initialized so that their hyperplanes should pass through the center of input pattern set, and those of output units are initialized to zero. Several simulations were performed to study how the initialization effects the structure forming process of the NN. From the simulation result, it was confirmed that the initialization gives better network structure and higher generalization ability.
  • GOTANDA Hiromu; SHIRATSUCHI Hiroshi
    Proc. IFAC2000 Symposium on System Identification, California (1) 49 - 54 2000/06 [Refereed]
  • H Gotanda; H Shiratsuchi; K Inoue; K Kumamaru
    (SYSID'97): SYSTEM IDENTIFICATION, VOLS 1-3 PERGAMON PRESS LTD 789 - 794 1998 [Refereed]
     
    This paper formulates a necessary condition for multilayer nets to have solutions by a set of normal vectors orthogonal to separation hyperplanes. Comparing the necessary condition to the distributions of normal vectors with the weights and biases initialized ordinarily by random numbers with zero mean, it is derived that bipolar nets are superior to unipolar nets in convergence of the back propagation learning initialized in such an ordinary manner.
  • GOTANDA Hiromu; SHIRATSUCHI Hiroshi; INOUE Katsuhiro; KUMAMARU Kousuke
    Journal of Robotics and Mechatronics 9 (5) 393 - 397 1997/10 [Refereed]
  • GOTANDA Hiromu; SHIRATSUCHI Hiroshi; INOUE Katsuhiro; KUMAMARU Kousuke
    Journal of Robotics and Mechatronics 9 (5) 398 - 405 1997/10 [Refereed]
  • GOTANDA Hiromu; SHIRATSUCHI Hiroshi; INOUE Katsuhiro; KUMAMARU Kousuke
    The Institute of Electronics, Information and Communication Engineers The Institute of Electronics, Information and Communication Engineers 80 (2) 696 - 699 0915-1923 1997/02 [Refereed]
     
    非線形ユニットがしきい値関数で与えられる場合について, XOR問題の解を分離超平面に直交する法ベクトルで定式化して, 解領域の形状が両極型と片極型ネットで異なることを示す. この結果からBP学習のアトラクタはネット極性に依存することが示唆される.
  • Studies on Cardinality of Solutions fot Multilayer Nets and Scaling Method in Hardware Implementations
    J. Robotics and Mechatronics 9 (5) 398 - 405 1997
  • GOTANDA HIROMU; SHIRATSUCHI HIROSHI; INOUE KATSUHIRO; KUMAMARU KOUSUKE
    IPSJ Journal Information Processing Society of Japan (IPSJ) 37 (8) 1535 - 1542 1882-7764 1996/08 [Refereed]
     
    This paper elucidates that multilayer nets of equal structure allow the same cardinarity of admissible solutions for learning tasks of which input patterns are related by Affine transform, even if their sigmoid functions are different in polarity and range. This result can be applied to a scaling problem arising on building the nets in analog hardware. In the case of the input patterns and the sigmoid functions multiplied by a scaling factor k, the separation and generalization abilities can be preserved if the weight values are set to 1/k times the original ones with keeping the bias value...
  • GOTANDA Hiromu; UEDA Yoshihiro; SHIRATSUCHI Hiroshi
    Journal of Robotics and Mechatronics 7 (4) 336 - 343 1995/04 [Refereed]
     
    本論文では多層ニューラルネットの解空間と学習挙動をユニット極性の観点から考察し，各層におけるユニット極性の組合せにより解の挙動を明らかにした．

Conference Activities & Talks

  • 小柳大輔; 白土浩; 五反田博; 久良修郭
    計測自動制御学会九州支部学術講演会予稿集  2013/11
  • 松崎隆哲; 白土浩; 久良修郭
    情報処理学会全国大会講演論文集  2013/03
  • アフリカ地域の学校教員を対象とした組込みシステム教育について  [Not invited]
    松崎 隆哲; 白土 浩; 久良 修郭
    平成25年度 情報処理学会第72回全国大会  2013/03  平成25年度 情報処理学会第72回全国大会
  • 田中宏典; 白土浩; 五反田博; 岩崎宣生
    計測自動制御学会九州支部学術講演会予稿集  2011/12
  • 周波数オフセット下でのOFDM伝送路推定  [Not invited]
    白土 浩; 五反田 博
    第30回計測自動制御学会九州支部学術講演会 (講演番号：202B3)  2011/12  大分  第30回計測自動制御学会九州支部学術講演会 (講演番号：202B3)
  • 岩崎宣生; 井上勝裕; 石橋孝昭; 松崎隆哲; 平野剛; 白土浩; 五反田博
    電気学会産業応用部門大会(CD-ROM)  2011/09
  • Tanaka Hironori; Iwasaki Nobuo; Shiratsuchi Hiroshi; Gotanda Hiromu
    Proceedings of the Society Conference of IEICE  2011/08
  • スパース尺度に基づく音環境のモニタリング  [Not invited]
    白土 浩; 五反田 博
    第29回計測自動制御学会九州支部学術講演会 (講演番号：104C3)  2010/12  宮崎  第29回計測自動制御学会九州支部学術講演会 (講演番号：104C3)
  • SimulinkによるOFDMシステムのシミュレータ開発  [Not invited]
    白土 浩; 五反田 博
    第29回計測自動制御学会九州支部学術講演会 (講演番号：104C4)  2010/12  宮崎  第29回計測自動制御学会九州支部学術講演会 (講演番号：104C4)
  • フィードフォワードループを有するディジタルサーボ系のサンプリング周波数に関する考察  [Not invited]
    白土 浩; 五反田 博
    第29回計測自動制御学会九州支部学術講演会 (講演番号：102A5)  2010/12  宮崎  第29回計測自動制御学会九州支部学術講演会 (講演番号：102A5)
  • SAYVISITH Vithaya; KIMURA Tetsuya; NAKAGAWA Fuminari; SHIRATUCHI Hiroshi; HARATANI Naomi; GOTANDA Hiromu
    IEICE technical report. Communication systems  2008/02 
    Carrier frequency offset (CFO) causes inter-carrier interference (ICI) between the sub-carriers in Orthogonal Frequency Division Multiplexing (OFDM) systems, which degrades the system performance. In this paper, a new approach based on independent component analysis (ICA) is proposed to cancel the ICI, to estimate the flat-fading sub-channel and to restore the transmitted symbols, by using only the received symbols. First, the ICI problem is formulated as an ICA model under the flat-fading sub-channels. Next, the permutation and the scaling indeterminacy inherent in ICA are resolved. Then t...
  • SAYVISITH Vithaya; KIMURA Tetsuya; NAKAGAWA Fuminari; SHIRATUCHI Hiroshi; HARATANI Naomi; GOTANDA Hiromu
    IEICE technical report  2008/02 
    Carrier frequency offset (CFO) causes inter-carrier interference (ICI) between the sub-carriers in Orthogonal Frequency Division Multiplexing (OFDM) systems, which degrades the system performance. In this paper, a new approach based on independent component analysis (ICA) is proposed to cancel the ICI, to estimate the flat-fading sub-channel and to restore the transmitted symbols, by using only the received symbols. First, the ICI problem is formulated as an ICA model under the flat-fading sub-channels. Next, the permutation and the scaling indeterminacy inherent in ICA are resolved. Then t...
  • ICAに基づくOFDM周波数オフセットの解消  [Not invited]
    五反田 博; 松本一良; 中河史成; 白土 浩
    2007/09
  • 並列レンダリングにおけるフレーム間差分情報を演算量低減について  [Not invited]
    五反田 博; 末松慶文; 白土 浩
    第15回電子情報通信学会九州支部学生会講演会論文集  2007/09  沖縄  第15回電子情報通信学会九州支部学生会講演会論文集
  • ブロック画素の差分演算に基づく遠隔PC画面監視システムの構築について  [Not invited]
    五反田 博; 飯塚大介; 白土 浩
    2007/09  沖縄
  • 高残響下でのブラインド信号分離  [Not invited]
    古屋武志; 五反田 博; 白土 浩; 金田圭市; 石橋孝昭
    2007/05  京都
  • iptablesを用いたネットワーク負荷分散システムの構築について  [Not invited]
    五反田 博; 飯塚大介; 白土 浩
    2006/09
  • 継続的認証機構を有するネットワーク利用申請システムの構築について  [Not invited]
    五反田 博; 前田雅浩; 白土 浩
    2006/09  宮崎
  • Xgridを用いた3Dレンダリングシステムの並列分散処理について  [Not invited]
    五反田 博; 末松慶文; 白土 浩
    2006/09
  • 独立成分分析に基づくサブキャリア間干渉の解消法について  [Not invited]
    五反田 博; 髙瀬成史; 中河史成; 白土 浩; 石橋孝昭
    2006/09  宮崎
  • SHIRATSUCHI Hiroshi; TOMORI Yoshimi
    IEICE technical report. Neurocomputing  2005/06 
    This report studies the coral health evaluation using a neural network, which is made to learn Ikema's water depth adjusting algorithms for extraction of the bottom features at optional points from remote sensing images. The IKONOS satellite images were used with high resolution of 82cm. Our previously proposed structural learning algorithm is applied to improve the generalization ability, since the training data, i.e., the measurements for correction of actual water depth are limited in number.
  • 衛星リモートセンシング画像によるサンゴ健全度評価に関する研究  [Not invited]
    白土 浩
    電子情報通信学会技術研究報告, NC 2005-22, pp.13-16  2005/06  電子情報通信学会技術研究報告, NC 2005-22, pp.13-16
  • テキスト・ナビゲータの提案  [Not invited]
    白土 浩
    第13回インテリジェント・システム・シンポジウム講演論文集, pp.128-129  2003/12  第13回インテリジェント・システム・シンポジウム講演論文集, pp.128-129
  • MIYAGI Tomokazu; SHIRATSUCHI Hiroshi
    IEICE technical report. Neurocomputing  2003/03 
    This report studies how our proposed initialization effects the rule extraction of neural networks by structural learning with forgetting. This proposed initialization consists of two steps: weights of hidden units are initialized so that their hyperplanes should pass through the center of input pattern set, and those of output units are initialized to zero. From simulation result on descovery of boolean function problem which has 5 and 7 inputs, it was confirmed that proposed initialization gives more simple network structure and higher rule extraction ability from improving the convergenc...
  • HIGA Naoki; SHIRATSUCHI Hiroshi
    IEICE technical report. Neurocomputing  2003/03 
    This report studies on an additional pruning algorithm with structure learning with forgetting(SLF) for the redundant units of neural networks. Two types redundancy criteria of hidden units are considered : differences between two hidden unit outputs, and using separation hyperplane's angle, norm and distance. Through the analysis of similarity progresses in SLF, in case of large number of initial hidden units, it is shown that the redundant hidden units are still remain at the end of learning. Proposal learning algorithm eliminates redundant the hidden units, which have the high similarity...
  • WATANABE Hiromasa; EZAKI Shu; GOTANDA Hiromu; SHIRATSUCHI Hiroshi
    IEICE technical report. ME and bio cybernetics  2003/01 
    A multichannel taste sensor composed of lipid/polymer membranes for transforming to electric potential outputs different patterns for substances with different taste qualities. Because of mutual interaction between taste substances, it was still difficult to decompose into the components from the response to the solution containing more than two taste dualities. In this paper, the estimation of the concentrations from the response to mixed taste solution is well done using a neural network system.
  • MIYAGI Tomokazu; SHIRATSUCHI Hiroshi
    IEICE technical report. Neurocomputing  2002/06 
    This report studies how our proposed initialization effects the structure determination of neural networks by structural learning with forgetting. This proposed initialization consists of two steps: weights of hidden units are initialized so that their hyperplanes should pass through the center of input pattern set, and those of output units are initialized to zero. From simulation result performed on iris classification problem, it was confirmed that proposed initialization gives more simple network structure and higher generalization ability.
  • HIGA Naoki; SHIRATSUCHI Hiroshi
    IEICE technical report. Neurocomputing  2002/06 
    This report studies how our proposed additional learning algorithm for structural learning with forgetting effects the structure determination of neural networks. The proposed learning algorithm eliminates redundant hidden units, which have the high similarity between hidden units, from neural networks. From simulation result performed on iris classification problem, it was confirmed that the proposed learning algorithm gives better network structure and higher generalization ability.
  • SHIRATSUCHI Hiroshi; SHIMABUKURO Kazuki; HIGA Naoki
    IEICE technical report. Neurocomputing  2002/03 
    This report studies that the Intrusion Detection System(IDS) using modular neural networks(MNN) improves the unknown attacks detection. It is proposed that the learning method for composing the personal profile at the normally situation on each module of the MNN. Each UNIX's commands with packet capturing are encoded by the characteristic attributes. These attributes are selected from among the behavior, which can guess the intruder, for example execute capability, using network ports, important keywords matching. From the simulation, using the actual network packet, it is shown that the pr...
  • HIGA Naoki; SHIRATSUCHI Hiroshi
    IEICE technical report. Neurocomputing  2001/06 
    In this report, a designed of neural network library for C language is studied. By expressing each parameter layout with hierarchy structure, each function requires only on structural argument which indicating the whole of neural nets. Even if complicated preprocessing stage such as the memory management or the structural design is not considered by using proposed library, the hierarchy neural network can be realized as a program. In addition, it is considered that expansion of nets structure and design of new function can be easily realize due to the proposed model has hierarchical structu...
  • SHIRATSUCHI Hiroshi; GOTANDA Hiromu; IKEMASU Yousuke; INOUE Katsuhiro; KUMAMARU Kousuke
    IEICE technical report. Neurocomputing  2001/03 
    In this report, it is studied that the method for recognition the time series signal by the multilayer neural net with preprocessing functions such as feature extraction and pattern normalization. The nets consist of modular subnets, each being specialized to preprocess its own signal and initialized to improve in signal recognition. As a simulation result, it is considered that the converging and recognizing performance of the nets is improved than ordinary method such as learning multiple signal as one net and using same preprocessing function for all signals.
  • SHIRATSUCHI Hiroshi; GOTANDA Hiromu; INOUE Katsuhiro; KUMAMARU Kousuke
    IEICE technical report. Neurocomputing  2000/03 
    This paper studies how our proposed initialization effects the structure determination of neural networks by structural leaning with forgetting. The proposed initialization consists of two steps: weights of hidden units are initialized so that their hyperplanes should pass through the center of input pattern set, and those of output units are initializedto zero. Then, in additional initialization which removes redundant hidden units at the outset of learning is described. From simulation result performed on iris classification problem and chaos prediction problem, it was confirmed that the ...
  • SHIRATSUCHI Hiroshi; GOTANDA Hiromu; SHIMADA Masaaki; INOUE Katsuhiro; KUMAMARU Kousuke
    IEICE technical report. Neurocomputing  1999/11 
    This paper studies how our proposed initialization effects the structure determination of neural networks by structural learning with forgetting. The proposed initialization consists of two steps: weights of hidden units are initialized so that their hyperplanes should pass through the center of input pattern set, and those of output units are initialized to zero. From simulation result performed on Iris problem, it was confirmed that the initialization gives better network structure and higher generalization ability.
  • SHIRATSUCHI Hiroshi; GOTANDA Hiromu; INOUE Katsuhiro; KUMAMARU Kousuke
    IEICE technical report. Neurocomputing  1998/10 
    This report studies an initialization of bipolar networks with single hidden layer trained for pattern classification problems. Weights of the hidden layer are initialized so that separation hyperpalnes should pass through the center of input pattern set, and those of the output layer are initialized zero. From simulation results for MONK's problem, Iris classification problem and Soner target identification problem, it is confirm that the proposed initialization gives better convergence in wider range of initial values and learning coefficent that the ordinal initialization that the weight...
  • SHIRATSUCHI Hiroshi; GOTANDA Hiromu; INOUE Katsuhiro; KUMAMARU Kousuke
    IEICE technical report. Neurocomputing  1997/11 
    This paper formulates a separability condition by normal vectors so that the separation hyperplanes should pass through a rectangle surrouding the input set. Then it depicts the distributions of two-dimensional normal vectors satisfying the condition. These distributions elucidate that the condition for the first hidden layer varies significantly even if the input patterns are simply translated, and that those conditions (in a wider sense) for the other layers are different between unipolar and bipolar units. It, also depicts an initial distribution of normal vectors with the weights initia...
  • GOTANDA Hiromu; SHIRATSUCHI Hirosh; KUMAMARU Kousuke; INOUE Katsuhiro
    IEICE technical report. Neurocomputing  1997/03 
    In this report, three conditions on weights and biases of hidden units in any layers are derived: The first is for their separation hyperplanes passing through their input set so as to match their active-regions with the input set; The secod is for fixing their activ-regions κ times as wide as the input set ; The third is for matching the network outputs with the targets. Three conditions are incorporated into a new initialization scheme of the back propagation algorithm. Simulation results on parity and sonar problems show that the initialization gives better convergence than usual random ...

MISC

Research Grants & Projects

  • Japan Society for the Promotion of Science：Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (C)
    Date (from‐to) : 2021/04 -2024/03 
    Author : 白土 浩
     
    本研究では大学等の高等教育機関における中途退学者の減少を目的に教育ビッグデータや学生が所有するスマートフォンやBYODによる構内無線LAN接続履歴や通信履歴を活用した学習指標として、「行動指標」「受講指標」「成績指標」を生成した上で深層学習により履修指導が必要な学生をスクリーニングすることを目的としている。 本年度は、まず「受講指標」を生成するために、先行研究において構築した学内ネットワーク統合システムへの機能拡張として、 Layer7 FireWall(L7FW) から取得できる通信ログ情報を格納させる方法について実装を進めた。具体的には、利用アプリケーションや接続先情報をもとに通信種別をカテゴライズされたデータベースとして構築する作業を進めた。次に、L7FW からは全セッションログが取得されるため、ログ解析およびデータベース化についてはバッチ処理として実装した。そして、「行動指標」としては構内無線 LAN コントローラおよび IEEE802.1x 認証装置からの各種ログ情報をもとに5分毎の滞在位置座標をデータベース化する機能について実装しているが、この情報を学内の位置情報を反映した画像情報として生成する方法について検討を進めた。さらに、「成績情報」としては学生の成績情報や受講状態・出欠情報等を集約する機能についてはコロナ禍でオンライン授業が増えたため収集が進んでいない状況である。また、従来とは授業形態が大きく変化し、仕様を再策定する必要もあると考えている。また、各指標を画像データとして可視化する点についても「行動指標」や「受講指標」が大学内での対面授業を前提とした仕組みであったため実装が進んでいない状況である。
  • Japan Society for the Promotion of Science：Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (C)
    Date (from‐to) : 2013/04 -2016/03 
    Author : GOTANDA Hiromu; SHIRATSUCHI Hiroshi; ISHIBASHI Taka-aki; MATSUZAKI Takanori
     
    In the research project, we have proposed a frame-wise DOA (Direction of Arrival) estimation using the sparsity of speech sound sources and a real-time sound source separation using soft-mask filters designed based on the DOA estimate. For fixed sound sources, our approach has been confirmed that the DOAs can be estimated with high accuracy and that the target sound can be separated in real time in high quality under the condition that the target sources are in a range of ±30 °, the reverberation time is not more than 200msec and the SN ratio is not less than 15dB. Under the same condition, it has been confirmed that the moving sound source can be tracked in real time with an accuracy of less than error 4 ° even if the sound moves in a variety of movement patterns at a speed of 32~96cm/s.
  • 日本学術振興会：科学研究費助成事業 奨励研究(A)
    Date (from‐to) : 2000 -2001 
    Author : 白土 浩
     
    本研究では階層型ニューラルネットの収束特性を解析するために,訓練データの正規化が学習収束にどのように影響するかを検討し,平均がゼロとなるように正規化した場合,著者らが先に提案した初期値設定法の中間層に対する初期値設定指針を自然に満たすため,初期値設定手順が簡略化されることを明らかにした.さらに,この正規化手順は学習終了後のネットの中間層荷重に簡単に埋め込むことができるため,認識対象データの正規化は不要となることを示した.アヤメ分類問題やMONK's分類問題,ソナー識別問題に対するシミュレーションの結果,提案法による場合,平均ゼロの一様乱数で荷重を初期化する通常の設定法に比べて,良好な収束結果が得られることを確認した.さらに,上述の初期値設定や正規化がネットの汎化能力に及ぼす影響についても考察し,両者を併用したときに最も良好な結果が得られることも確認した. 正規化や特徴抽出などの前処理過程を埋め込んだ多層ネットで時系列信号を識別させる方法について,複数のサブモジュールから構成されるモジュール構造ネットを用い,サブモジュール毎に最適な前処理を個別に埋め込む方法を提案した.そして,睡眠脳波信号に対する認識実験の結果,通常のように複数の信号を1つのネットで学習させたり,全ての信号に同一の前処理をした場合に比べて良好な認識結果が得られた.また,実際の睡眠脳波信号系列をオンライン提示することで,対象信号のセグメンテーションが可能となることを示した. 以上による成果を簡単に利用できるようにC言語用のライブラリ形式で公開し,初心者でも容易に操作できるようGtk+を使ったGUI環境も作成した.

Other link

researchmap


Copyright © MEDIA FUSION Co.,Ltd. All rights reserved.