Updated on 2024/09/26

写真a

 
KANMA Koji
 
Name of department
Faculty of Systems Engineering, Intelligent Informatics
Job title
Lecturer
Concurrent post
Informatics Division（Lecturer）
External link

Education

  • 2018
    -
    2021

    Wakayama University   Graduate School of Systems Engineering  

  • 2012
    -
    2014

    The University of Tokyo   The Graduate School of Engineering   Department of Chemical System Engineering  

  • 2007
    -
    2012

    The University of Tokyo   The Faculty of Engineering   Department of Chemical System Engineering  

Academic & Professional Experience

  • 2022.07
    -
    Now

    Wakayama University   Faculty of Systems Engineering   Project Assistant Professor

  • 2020.04
    -
    2022.04

    Japan Society for the Promotion of Science   特別研究員

Association Memberships

  • IEICE

Research Areas

  • Informatics / Perceptual information processing / Machine Learning

  • Informatics / Perceptual information processing / Computer Vision

Research Interests

  • machine learning

  • deep learning

Published Papers

  • Pruning based on Activation Function Outputs for Efficient Neural Networks

    Koji Kamma, Toshikazu Wada （Part： Lead author )

    2023 IEEE International Conference on Systems, Man, and Cybernetics (SMC)     2023.10

  • Pruning with Output Error Minimization for Compressing Deep Neural Networks

    Koji Kamma, Toshikazu Wada

    MIRU2023     2023.07

  • Pruning Ratio Optimization with Layer-Wise Pruning Method for Accelerating Convolutional Neural Networks.

    Koji Kamma, Sarimu Inoue, Toshikazu Wada

    IEICE Transactions on Information & Systems   105-D ( 1 ) 161 - 169   2022

    DOI

  • REAP: A Method for Pruning Convolutional Neural Networks with Performance Preservation.

    Koji Kamma, Toshikazu Wada

    IEICE Transactions on Information & Systems   104-D ( 1 ) 194 - 202   2021

    DOI

  • Neural Behavior-Based Approach for Neural Network Pruning.

    Koji Kamma, Yuki Isoda, Sarimu Inoue, Toshikazu Wada

    IEICE Transactions on Information & Systems   103-D ( 5 ) 1135 - 1143   2020

    DOI

  • Reconstruction Error Aware Pruning for Accelerating Neural Networks.

    Koji Kamma, Toshikazu Wada

    Advances in Visual Computing - 14th International Symposium on Visual Computing ( Springer )    59 - 72   2019

    DOI

  • Behavior-Based Compression for Convolutional Neural Networks.

    Koji Kamma, Yuki Isoda, Sarimu Inoue, Toshikazu Wada

    Image Analysis and Recognition - 16th International Conference ( Springer )    427 - 439   2019

    DOI

  • Development of a Novel Spectra Analysis Method to Construct Accurate NIR Models

    Kamma Koji, Kaneko Hiromasa, Funatsu Kimito

    Journal of Computer Aided Chemistry ( Division of Chemical Information and Computer Sciences The Chemical Society of Japan )  15   1 - 9   2014

    　View Summary

    Near-infrared spectroscopy (NIR) is widely used for non-destructive food quality check. The prediction models are constructed between NIR spectra and quality parameters. However, because of the noise included in spectra and the duplication between the peaks of the target components and those of the other components, the prediction accuracy of the models decreases. To avoid this problem, derivative spectra are used in modeling. Derivation of spectra has an effect to emphasize the small and narrow peaks so that the affection of peak duplication decreases. On the other hand, derivation of spectra also has an effect to enlarge the noise. The impacts of these effects change as the derivative changes, hence it is necessary to select the adequate derivative for each data. Besides, if there are several peaks of the target components, the adequate derivative is different for each peak. In this paper, we therefore construct regression models using the spectra, the first, second and third derivative spectra, and the combinations of them. The accuracy of the models which are constructed with different derivative spectra or the combinations of them changes when the number of the training data changes. Thus, we proposed a method to select the proper model according to the number of the training data. The selection is performed based on the prediction accuracy of each model. A simulation data set that mimics the spectra where three different peaks duplicate was analyzed using the proposed method. Then, the proposed method was applied to sugar content prediction of oranges. The results showed that the most accurate model changed as the number of the training data changed, and that the effectiveness of the proposed method was proved.

    DOI

  • Construction of two-dimensional quantitative structure retention relationship models and structure elucidation based on inverse analysis with QSRR models

    Kamma Koji, Kaneko Hiromasa, Funatsu Kimito

    Proceedings of the Symposium on Chemoinformatics ( Division of Chemical Information and Computer Sciences The Chemical Society of Japan )  2013   O10 - O10   2013

    　View Summary

    Gas chromatography (GC) and two-dimensional GC (GC-GC) are widely used for separation, structure elucidation and quantitative analysis. In GC and GC-GC, the chemical structure is elucidated by comparing the measured retention time of each compound with the database. But, structure elucidation is infeasible if the retention time is not available from the database. Thus, quantitative structure retention relationship (QSRR) is proposed to predict the retention time from the structure. Some researchers constructed the QSRR models specialized for the limited types of compounds. In this study, we aim to construct the QSRR models that can predict the retention time of various compounds in GC-GC with high accuracy. In addition, we propose a structure elucidation method based on the inverse analysis with the models. First, the objective value of the retention time is set. Then, structure elucidation is accomplished by comparing the objective value with the predicted retention time of new structures. The prediction errors can be a problem in comparison between the predicted and objective values. We deal with this problem by setting an acceptable error for each compound based on the reliability of the predicted value. The analysis with the measurement data proved the effectiveness of the proposed method.

    DOI

  • Proposal of a novel near-infrared spectral analysis method for constructing robust and high-precision models

    Kamma Koji, Kaneko Hiromasa, Funatsu Kimito

    Proceedings of the Symposium on Chemoinformatics ( Division of Chemical Information and Computer Sciences The Chemical Society of Japan )  2012   1B1a - 1B1a   2012

    　View Summary

    Non-destructive testing of food quality with Near-Infrared Spectroscopy (NIR) is becoming common. The prediction models are constructed between NIR spectra and quality parameters. Many investigations have been done for the construction of high predictive models. Although some models indeed have suitably predictive accuracy, those models work well in only limited data domains and the accuracy decreases with time. Hence the models should be reconstructed with new data by wasting samples of objective foods and measuring the quality. To perform both the reduction of the loss of the food and the high performance of the models, overlapped peaks of NIR spectra should be considered because the overlapped peaks make relationships between NIR spectra and quality parameters unclear. Derivation of spectra is generally used to solve this problem. An adequate order of derivative changes depending on how peaks are overlapping, but the dependence of an adequate derivative order on the number of training samples remains to be clarified. Therefore we propose a method using some kinds of derivative order of spectra according to the number of samples for the construction of regression models. The effectiveness of the proposed method was confirmed thorough the analyses of simulation data and real data.

    DOI

▼display all

Awards & Honors

  • PRMU研究会ベストプレゼンテーション賞

    2022.12    

  • 情報処理学会2020年度研究会推薦論文

    2021.05    

  • ICIAR2019 best paper award

    2019.08    

  • MIRU学生奨励賞

    2019.07    

Conference Activities & Talks

  • DNN モデルの圧縮法 - 効率的な AI の実現に向けて

    菅間幸司

    電子情報通信学会総合大会  2023.03.07  

Patents

  • ニューラルネットワーク処理装置、ニューラルネットワーク処理方法、及びコンピュータプログラム

    Patent no： 特許第7533933号

    Date registered： 2024.08.05 

    Date applied： 2020.07.20 （ 特願2020-123973 ）   Publication date： 2022.02.01 （ 特開2022-20464 ）  

    Inventor(s)/Creator(s)： 和田俊和、菅間幸司  Applicant： 国立大学法人和歌山大学

  • ニューラルネットワーク処理装置、コンピュータプログラム、ニューラルネットワーク製造方法、ニューラルネットワークデータの製造方法、ニューラルネットワーク利用装置、及びニューラルネットワーク小規模化方法

    Patent no： 特許第7438544号

    Date registered： 2024.02.16 

    Date applied： 2019.08.28 （ 特願2020-546831 ）   Public disclosure date： 2020.03.19 （ WO2020/054402 ）

    Inventor(s)/Creator(s)： 和田俊和、菅間幸司、磯田雄基  Applicant： 国立大学法人和歌山大学

  • ニューラルネットワークの圧縮方法、ニューラルネットワーク圧縮装置、コンピュータプログラム、及び圧縮されたニューラルネットワークデータの製造方法

    Patent no： 特許第7438517号

    Date registered： 2024.02.16 

    Date applied： 2019.07.25 （ 特願2019-137019 ）   Publication date： 2021.02.18 （ 特開2021-022050 ）  

    Inventor(s)/Creator(s)： 和田俊和、菅間幸司  Applicant： 国立大学法人和歌山大学

  • ニューラルネットワークの圧縮のためにコンピュータによって実行される方法、ニューラルネットワーク圧縮装置、コンピュータプログラム、及び圧縮されたニューラルネットワークデータの製造方法

    Date applied： 2024.06.14 （ 特願2024-096813 ）  

    Inventor(s)/Creator(s)： 菅間幸司  Applicant： 国立大学法人和歌山大学

  • 画像処理装置

    Date applied： 2022.12.07 （ 特願2022-195687 ）  

    Inventor(s)/Creator(s)： 和田俊和、菅間幸司、岡山敏之、野口　威、島田佳典 

KAKENHI

  • DNNアーキテクチャの対称性破れを利用した効率的学習法の開発

    2024.04
    -
    2026.03
     

    Grant-in-Aid for Early-Career Scientists  Principal investigator

Joint or Subcontracted Research with foundation, company, etc.

  • AIを利用した画像解析によるミニトマトの生育診断技術の開発

    2024.06
    -
    2025.02
     

    Joint research  Co-investigator

  • 商品棚画像の解析及び精度向上に関する研究

    2024.04
    -
    2025.03
     

    Joint research  Co-investigator

  • Deep Learningを用いた光学検査装置の新規ソリューションの開発

    2024.04
    -
    2025.03
     

    Joint research  Co-investigator

  • 画像解析によるミニトマトの生育診断手法の開発

    2023.04
    -
    2024.02
     

    Joint research  Co-investigator