Work in knowledge-based systems in Japanese industrial laboratories appears to be tightly coupled with application or product development. Japanese computer manufacturing companies and certain high tech companies carry out some knowledge-based systems research. Other types of companies do virtually none. The JTEC team observed a thin layer of excellent industrial research at Fujitsu, Hitachi, Toshiba, NEC, and NTT. From other publications, we know that there is excellent knowledge-based system work at IBM Japan and Sony. Perhaps the most extensive and the deepest of the basic research activities at companies was seen at Hitachi's Advanced Research Laboratory (under the direction of Dr. Hiroshi Motoda), at NEC, and at NTT.
The JTEC panel surveyed topics covered by industrial researchers of knowledge-based systems in Japan by looking at three years' worth of volumes of working papers of the special interest group on knowledge-based systems of the Japan Society for Artificial Intelligence. We looked at titles and abstracts only. Most of the work appeared to be follow-up studies to work that had entered the literature in the United States or Europe. The interest among Japanese researchers was very broad, touching many of the issues and research topics of current interest in the United States:
As to research on large knowledge bases, NEC demonstrated systems containing extensive data on DNA and protein sequences, and on protein secondary structure. The systems are used for searching for common patterns in the sequence information. In other areas of AI, NEC has a well-documented body of research on model-based diagnosis, genetic information processing, learning (using a technique called Minimum Description Length), learning theory, inductive learning, knowledge acquisition for classification problems and for consultation systems, natural language understanding, and case-based reasoning. The company believes that the important aspects of second generation ES tools are the incorporation of machine learning, model-based reasoning, cooperative problem solving, and facilities for extracting knowledge from large databases.
NEC has also developed a language, called PRIME (PRimary Inference Mechanism with Environment), for describing a semantic model of a domain. The model can be used for deep-level inference. The inference process can then be compiled into a shallow knowledge base.
Fujitsu Laboratories is conducting research in constraint satisfaction and optimization (Maruyama, Minoda et al. 1991), object-oriented knowledge bases for engineering, and machine learning. Mr. Maruyama presented a novel approach to solving constraint satisfaction and optimization problems, which he considers more effective than integer programming. The object-oriented work is aimed toward flexible management of design data, intelligent CAD support (including management of design constraints and constraint-based animation), and advanced methods for management of engineering information. Machine learning research has focused on applying computational learning theory toward inductive inference of decision trees in the presence of noisy data.
AI research at NTT is conducted in three major areas: common-sense reasoning (in a restricted sense); machine translation; and VLSI design. The work on common-sense AI focuses on making quantitative judgments in very large KBs. For example, how long numerically is a long river? The research also includes judging ambiguity in words -- a long vacation vs. a long pencil. The machine translation work is based on a belief in the great importance of specialized knowledge. Currently the system has 15,000 sentence structure descriptions. The VLSI design research is aimed at automatic synthesis techniques from high level descriptions of circuits (see Rich 1992).