Welcome to Takao Mohri's Home Page

• Personal Information
  • I'm now a post-doctral Student at the University of Tokyo.
  • I belongs to Hidehiko-Tanaka Lab. in Department of Electric.
  • Postal address:
    Room 301, Building #13, Hidehiko-Tanaka Lab.
    Dept. of EE., 7-3-1 Hongo Bunkyo-ku, Tokyo 113 JAPAN.
  • Telephone:
    +81-3-3812-2111 ext. 7413
  • Fax:
    +81-3-5800-6922

• Research Activities

  My major research topic is 'Memory-Based Reasoning' in Artificial Intelligence. In the case of memory-based reasoning, the system remembers a large number of experiences, which consist of pairs of problems and solutions. The system solves problems by searching the most 'similar' problem from its storage, and applies the past solution to the current one.

  My research interest is how to define this similarity, and to apply memory-based reasoning to the weather forecasting, which is a big and practical application.

• Published Papers (in English)

  Title:

  An Optimal Weighting Criterion of Case Indexing for Both Numeric and Symbolic Attributes
  

  Authors:

  Takao Mohri and Hidehiko Tanaka
  

  Published-at:

  AAAI-94 Case-Based Reasoning Workshop, 1994, pp123-127
  Camera-ready: (PostScript, gnuziped,43926 bytes)

  Abstract:

  Indexing of cases is an important topic for Memory-Based Reasoning(MBR). One key problem is how to assign weights to attributes of cases. Although several weighting methods have been proposed, some methods cannot handle numeric attributes directly, so it is necessary to discretize numeric values by classification. Furthermore, existing methods have no theoretical background, so little can be said about optimality.
  We propose a new weighting method based on a statistical technique called Quantification Method II. It can handle both numeric and symbolic attributes in the same framework. Generated attribute weights are optimal in the sense that they maximize the ratio of variance between classes to variance of all cases.
  Experiments on several benchmark tests show that in many cases, our method obtains higher accuracies than some other weighting methods. The results also indicate that it can distinguish relevant attributes from irrelevant ones, and can tolerate noisy data.
  

  Title:

  Weather Forecasting Using Memory-Based Reasoning

  Authors:

  Takao Mohri, Masaaki Nakamura and Hidehiko Tanaka
  

  Published-at:

  Second International Workshop on Parallel Processing for Artificial Intelligence, 1993
  Camera-ready: (PostScript, gnuziped, 79500 bytes)
  

  Abstract:

  We have implemented a weather forecasting system, WINDOM, which uses memory-based reasoning. The observation data from Japan Meteorological Agency's networks were used directly as input data and the system was used to predict the weather. We used 5 categories such as clear, clouded, rainy, snowy, stormy, or 2 categories such as rainy or not rainy, to describe Tokyo 6 hours ahead. Two weighting methods of features, per category importance and cross category importance, were tested. The per category importance method, the more precise one, turned out to be too sensitive for the proportion of the answer categories and performed worse than em cross category importance} method. In the experiments, a hit rate of 84.1% was achieved, where the task was to predict if the weather 6 hours ahead would be rainy or not rainy.