Welcom to Katsunari Shibata's Home Page

• Personal Information
  Birthday: 1964.12.16 Sex : Male
  Affiliation: Dept. of Electrical and Electronics Engineering
  Faculty of Engineering, Oita University
  Position: Associate Professor

  Location / Telephone number

  Room No.420, on the 4th floor in the building of Electrical Class

  TEL: +81-97-554-7832(Direct), or Ext.7832

  Research Target
  A robot whose knowledge is not given from a human as a designer, but is obtained by itself through its experience as much as possible.
  In the case that a robot looks intelligent by some given knowledge:
  the designer is very intelligent, but the robot itself is not intelligent actually.
  In other words, persuit why our human are so intelligent than the modern intelligent robots is my research theme. This is a so deep research target that it may be impossible to understand the mechanism completely in my life.

  The recent reserch theme

  1. Basic Idea
     Let's make a robot with real intellingece !

     << Keyword 1 >>
     Turn from given knowledge to acquired knowledge
     I think that it is necessary to turn our idea about intelligent robots oppositely in the direction of how the given knowledge can be reduced. Reinforcement learning is the best solution to learn by itself.

     << Keyword 2 >>
     Turn from division of functional modules to growing
     In the present Robot and Brain research -> Division into functional modules (such as recognition, memory, planning, control)
     The viewpoint what the whole individual should do is required. -> a neural network from sensors to motors
     Growing -> from simple structure and easy tasks.

     << Keyword 3 >>
     Extension of Reinforcement learning:
     action learning --> learning for the whole process from sensors to motors
     After that, necessary functions are emerged in the neural network according to necessity.

     A typical research fruit at present:
     Box pushing behavior is learned without any given knowledge about image processing, control, or task.
     

  2. Basic Learning Algorithm
     [1]Temporal Smoothing Learning
     [2]Correlated Information Extracting Learning

     Acquisition of Functions by Reinforcement Learning

     * Direct-Vision-Based Reinforcement Learning

     * Learning of Hand Reaching and Hidden Representation

     * Actor-Q Architecture and Learning of Active Perception

     *Learning of Selective Attention and Adaptive Formation of Associative Memory

     *Learning of Prediction

     * Learning of Memory

     *Learning of Communication for Negotiation and Emergence of Individuality

     *Emergence of Symbol (Discrete representation) using a Neural Network

     *Emergence of Individuality and Sociality

     *Learning of Reward Distribution

     Sensor Integration Using Neural Networks

     * Integration of Many Local Sensor Signals like Retina and Some Models of Visual System using [1]

     * Multiple Sensor Integration using [2]

     Recurrent Neural Network

     *Simple Learning Algorithm of Recurrent Neural Networks

     *Selective Attention and Adaptive Formation of Associative Memory

     *Learning of Communication for Negotiation and Emergence of Individuality

     *Learning of Prediction

     Architecture and Hidden Representation of Neural Network

     *Gauss-Sigmoid Neural Networks

     *Space Reconstruction and Hidden-level Generalization

     *Growing Neural Network

     Multi Agent System

     *Emergence of Individuality and Sociality by Reinforcement Learning

     *Learning of Communication for Negotiation and Emergence of Individuality

     *Learning of Reward Distribution

     Growth

     *Body Growth Effect in Reinforcement Learning

     *Growing Neural Network

  Publication List

  The theme of Master Thesis

  "The Learning ability of the robot using Back Propagation Method"

  1989.3 Inoue-Inaba Lab. Mechanichal Eng. Univ. of Tokyo

  The theme of Doctor Thesis

  "Autonomous Learning System Using Neural Networks"

  1997.3 Okabe-Hirose Lab., RCAST, Univ. of Tokyo

  Acacdemic Society

  JNNS, IEEE, IEICE

  Former Affiliations

  Nagao & Mitsuishi Lab., Dept. of Mechanical Engineering, Univ. of Tokyo

  Inoue & Inaba Lab., Dept. of Mechanical Engineering, Univ. of Tokyo

  Okabe & Hirose Lab., Research Center of Advanced Sci. & Tech., Univ. of Tokyo

  Ito Lab., Dept. of Comp. Intelligence & Sys. Sci., Tokyo Inst. of Tech.