Kazuma Matsumoto

Research Theme

keywords: Learning Classifier System, Deep Learning, Data Mining, Genetic Algorithm, Reinforcement Learning

As the accuracy of deep learning improves, deep learning is used for various purposes recently. Deep learning can cope with high-dimensional input, and its accuracy of classification overcomes human in some fields. However, since knowledge gained from deep learning is difficult for humans to understand, it hardly uses knowledge gained by deep learning. On the other hand, Learning Classifier System (LCS) is hybrid machine learning of reinforcement learning and evolutionary computation, which can gain generalized if-then rules which are easy for human to understand. However, it is difficult for LCS to learn from high-dimensional input. This research proposes hybrid system of LCS and deep learning to obtain generalized if-then rules which are easy for human to understand from high-dimensional input.

Papers, International Conferences

  • Kazuma Matsumoto, Takato Tatsumi, Hiroyuki Sato, Tim Kovacs, Keiki Takadama:

    “XCSR Learning from Compressed Data Acquired by Deep Neural Network”,
    Journal of Advanced Computational Intelligence and Intelligent Informatics. 21. 856-867. 10.20965/jaciii.2017.p0856. (2017)

  • Kazuma Matsumoto, Rei Saito, Yusuke Tajima, Masaya Nakata, Hiroyuki Sato, Tim Kovacs, Keiki Takadama:

    “Learning Classifier System with Deep Autoencoder”,
    IEEE Congress on Evolutionary Computation (CEC), July 2016, Vancouver, CANADA.

Languages Skills

Japanese, C, C++, C#, Java, Ruby, Fortran, HTML, CSS, XML, LaTeX, etc.