« iepriekšējāTurpināt »
descriptions of research activities, is personally interested in large sys- that program. By using the fork and and RCAST Journal (once a year), tems, for example the large knowl- join mechanism in Unix, it forks a which contains outlines and summa- edge base that is needed by the Elec- new process and runs the program, ries of research, lists of publications, tronic Dictionary Research project eventually returning either 'yes' or and current research status. General (see "edr.92", 27 March 1992); how- 'no' according to the end code of the questions about RCAST can be ad- ever, he feels that such a knowledge program. Thus many Unix programs dressed to
base can't be flat, i.e., it needs to be can mesh nicely with KAUS, includ
built as a multilevel structure. Par- ing various database systems. Prof. Setsuo Ohsuga, Director enthetically, the question of ICOT Ohsuga and his group have Research Center for Advanced was brought up. Ohsuga felt that developed a variety of applications Science and Technology,
their hardware was okay, but that using KAUS, primarily in engineerThe University of Tokyo there were no substantial applica- ing domains, including mechanical 4-6-1 Komaba
tions developed yet. He also echoed design, feedback control system Mejuro-ku, Tokyo 154 JAPAN the Japanese sentiment that ICOT analysis, airplane wing design, chemiTel: +81 3-3481-4411
had been a very good advertisement cal compound design. Some details
for AI, and in fact had moved stu- and references are described below. Interestingly, Prof. Ohsuga also dents away from traditional areas, to What is important about this work, holds the chair of Knowledge pro- the point now that the best students in my opinion, is the very close link cessing within the Advanced systems in Japan seem to be moving into AI between advanced knowledge prodepartment, the focus of the current subfields. He also offered that ICOT cessing concepts and traditional engireport (this dual appointment speaks had propelled the growth of the AI
had propelled the growth of the AI neering issues. Most of the students worlds about the administrative in Japanese industry.
working in Ohsuga's lab are trained structure of this research organiza
in computer science, not engineering. tion). OHSUGA LAB
Nevertheless, the lab as a whole has I have reported on RCAST
a strong bias toward practical probactivities in the past, see for example Ohsuga has proposed a knowl- lems as seen by the applications they ["tachi.lab", 9 March 1992). Professor edge representation language (called are studying. RCAST's philosophy Tachi is in the Advanced Devices multilayered logic) based on an ex- emphasizes openness. When it is department, and under the chair of tension of predicate logic by expand- coupled to the quality of the work Intelligent Sensing Devices.
ing its syntax. (Ohsuga believes that that is occurring there, collaborative Ohsuga is a former President of knowledge representation is the activities with Western scientists are the Japan Society for Artificial Intel- central issue in Al. He also feels that bound to be fruitful. ligence and has been engaged in multilayered logic satisfies most re- A special interest at this lab is research in Al since 1970. His lab is quirements for knowledge represen- the intelligent CAD systems, i.e., the jointly run with Koichi Hori who tation; it is very convenient for defin- development of knowledge-based joined in 1988 from the National ing structured objects, and is also CAD system. In one feedback conInstitute of Japanese Literature convenient to describe and to link to trol system application, the system (Email: HORI@OHSUGA.RCAST. databases and existing procedures.) offers the designer a diagram lanU-TOKYO.AC.JP). Currently, a The Lab has also implemented a guage to define an object model of a professor from France and 19 stu- knowledge base system (KAUS; control system with block diagrams dents are visiting; about one third Knowledge Acquisition and Utiliza- and mathematical expressions. Roufrom outside Japan. Although the tion System) based on this language. tine tasks such as deriving transfer working language is Japanese, Eng- KAUS runs under Unix. It uses functions and calculating time relish is widely spoken and most of the a special procedural atom, 'Exec', sponses, are mostly done automatitechnical reports are available in that allows cooperation between cally. The system is distributed by English. Computer user-interfaces knowledge processing and conven- using various computing facilities on are mostly in Japanese however. tional computing in a simple way.
a network. Such feedback control On the general topic of expert Exec is a second order predicate that system design involves two kinds of systems, Ohsuga pointed out that accepts character strings as argu- man-machine cooperative activitthese are still fairly small scale in ments. It interprets the first argu- ies-one to build the control system Japan, but the systems are being ment as a program name and the model in the computer, and the rapidly implemented by industry. He remaining arguments as arguments of other to determine various physical parameters based on the model (time One of the most interesting models can directly be converted to responses, frequency responses, root papers on this subject appeared in procedural programs. He is building locus, relative stabilities, etc.). Computer Aided Design, Vol 21, No. a system using the meta-level mecha
We were also told about the 5, June 1989, pp. 315-337, Toward nism of KAUS to explore these prototype aircraft wing design system Intelligent CAD Systems, S. Ohsuga. I ideas. developed by using KAUS. Ohsuga recommend it to everyone that is Prof. Hori is involved in the pointed out the difficulties associated studying this topic as a readable and articulation problem. This is the with such designs, and that it may thought provoking summary.
process of cutting a set of symbols require as many as ten years for Another related system is Chem- out of the nebulous mental world. "I complete design and manufacture, ilog. This is a logic programming have a concept but I can't verbalize and that it may require not only language and system to support the it." It is necessary to articulate this numerical and large scale database development of chemical knowledge mental world in order to acquire processing but also knowledge pro- information processing systems. It is knowledge, model information, etc. cessing and model processing. The an extension of Prolog and is viewed Prof. Hori and coworkers have built prototype system has two numerical as a constraint logic programming a system, AA1 for aiding this articupackages, one for aerodynamic and language. In Chemilog, graphs repre- lation. Experts often put fragments the other for structural analysis, but senting chemical structures are treat- of ideas consisting of several words other packages can obviously be add- ed as basic components and isomor- and hand-drawn figures on paper, ed. A database of NACA wing sec- phic graphs and are identified. Con- and frequent erasure and rewriting tions is stored as standard wing sec- straints such as substructure recogni- are added when thinking about new tions and can be referred to by de- tion and replacement of substruc- ideas. Hori claims that AA1 assists in signers, or they can add their own tures make it possible for users to this stage, and they have experimentwing sections to the database. A wing write programs that handle chemical ed with it in the conceptual design of model consists of an aerodynamic structures in a visual form. There are automobiles. Currently they are and structural model. In aerodynam- two parts, an inference engine and a trying to investigate, from the cogniic design, designers first make an chemical structure database. Ohsuga tive science point of view, why the initial model that specifies the shape emphasizes that while this project is system works so well. of the wing, then they analyze and in the domain of chemistry the meth- Mainly, Hori's work concerns modify this, repeatedly, until goal od can also be applied in other engi- the bottom-up symbolization process characteristics such as lift, drag, and neering domains where graph repre- in the human mind. Ohsuga's work is moment are obtained. In structural sentation is used. CHAUS is an concerned with solving a problem design, designers first make an initial extension of Chemilog using KAUS. symbolically modeled. Currently, they model that specifies the wing frame- This was a national project that en- are trying to connect these two activwork constructed by beams and ribs, listed the aid of chemists throughout ities. Ohsuga emphasized that he then they analyze and modify it until Japan who prepared knowledge sheets also has interests in human interface, the wing has enough strength. In describing what functionality appears natural language understanding, case it isn't possible to obtain a when certain kinds of structural analogical reasoning, abductive reastrong enough structural model, the changes in chemical compound are soning, nonmonotonic reasoning, aerodynamic model must be modi- made.
learning systems, hypothesis generafied. Since both models are integrat- Ohsuga has recently been think- tion, etc., with special emphasis on ed into this system, consistency can ing about applying KAUS to the the world of engineering. be maintained. Associated with this problem of software design. He project is a package, MODIFY claims that an important step in (Modeling tool for Integrated Finite software design is conceptual model
Recent Papers element analysis), developed in the ing (rather than formal specificalab. This is an object oriented FEM tions) and that knowledge processing S. Ohsuga, "How Can Knowledge system, which uses parts objects, and technology can be fully used in the Based System Solve Large Scale the concept of degree of importance process of conceptual modeling.
Problems—Model Based Decomposito help with mesh generation. Moreover he claims that conceptual
tion and Problem Solving," to appear in Knowledge Based Systems, 5(3), Kousuke Yoshizawa, "Soft-Defi1992.
ned Machine" Satoshi Kobayashi, "A
Method for Acquiring Problem DeK. Hori and S. Ohsuga, "Word Space composition Strategy from Traces" Processor for Assisting the Articula
Einoshin Suzuki, "Framework for tion of the Mental World," in Infor- Connecting Several Knowledge-Bamation Modelling and Knowledge
sed Systems Under a Distributed Bases, Ohsuga, Kangassalo, Jaakkola,
Akira Utsumi, "Primitive Based
Representation of Adjectives and
Figurative Language Understanding" S. Ohsuga, "Framework of Knowledge-Based Systems-Multiple meta- Nigel Ward, "Structured Connelevel architecture for representing ctionist Studies in Natural Language problems and problem-solving pro- Processing" cesses," in Knowledge Based Systems, 3(4), 204-214 (1990).
Ionnis Zannos, "Interactive Mu
sic Generation System" T. Akutsu, E. Suzuki, S. Ohsuga,
Jari Vaario, "Study on the De"Logic-based Approach to Expert
velopment of Neural Networks in the Systems in Chemistry," in Knowledge
Context of Artificial Life" Based Systems, 4(2), 103-116 (1991).
During the JTEC visit to Shinichiro Yano, "An Approach RCAST, Prof. Ohsuga also provided to Aid Large-Scale Design Problem us with a package of preprints from by Computer" graduate students working in his lab. The titles and authors of these pa- Y. Minagawa, "The Study to Aid pers are listed below, but for specific Finite Element Method's Preprocessinformation it would be best to con- ing with Knowledge Engineering" tact Ohsuga or Hori directly.
Hiroyuki Yamauchi, "KAUS: Akira Tajima, "Supporting the Knowledge Acquisition and Utiliza- Development of Intelligent CAD tion System"
System on KAUS"
David K. Kahaner joined the staff of the Office of Naval Research Asian Office as a specialist in scientific
computing in November 1989. He obtained his Ph.D. in applied mathematics from Stevens Institute of Technology in 1968. From 1978 until 1989 Dr. Kahaner was a group leader in the Center for Computing and Applied Mathematics at the National Institute of Standards and Technology, formerly the National Bureau of Standards. He was reponsible for scientific software development on both large and small computers. From 1968 until 1979 he was in the Computing Division at Los Alamos National Laboratory. Dr. Kahaner is the author of two books and more than 50 research papers. He also edits a column on scientific applications of computers for the Society of Industrial and Ap. plied Mathematics. His major research interests are in the development of algorithms and associated software. His programs for solution of differential equations, evaluation of integrals, random numbers, and others are used worldwide in many scientific computing laboratories. Dr Kahaner's electronic mail address is: firstname.lastname@example.org. tokyo.ac.jp
Chunye Li, "Problem Model Design/Transformation-Based Program Development"
Masanori Sugimoto, "A Method to Assist the Acquisition and Expression of Subjective Concepts and Its Application to Design Problems"
Ning Zhong, "Knowledge Discover and Management in Integrated Use of Knowledge-Bases and Databases"
Yasunuki Sumi, "A Study of
HITACHI ACTIVITIES IN AI
AI and Expert System activities at Hitachi are reviewed.
Current and research projects are mentioned.
David K. Kahaner
In March 1992, a JTEC (Japa- X-Windows and Motif, which is NEL 2 these are implemented in a nese Technical Evaluation Center) claimed to be very user friendly.) hierarchical structure via
an team, headed by Prof. Edward More than 4,000 copies of ES/KER- object-oriented representation. PriFeigenbaum, Stanford University NEL have been sold in Japan. vate memos represent unstructured [Email: FEIGENBAUM@SUMEX- Source was written in C. Users can knowledge, suitable for the descripAIM.STANFORD.EDU), visited make use of various reasoning strate- tion of transient facts that may be Japan to assess and report on Japa- gies, including the following: obtained in the course of reasoning. nese activities in the areas of expert
ES/KERNEL 2 also comes with systems (ES). An oral summary of • Production system: Reasoning a graphical knowledge editor as well their report was presented in June proceeds through repetitive as a knowledge tailor, which allows 1992 at the NSF in Washington, DC,
sequences of search for
knowledge to be moved from one and the written report will be avail- selection and execution of representation to another. In addiable soon. The JTEC team visited a applicable rules.
tion, there are two domain shells, large number of laboratories at Uni- • Object oriented: Reasoning ES/Promote/W-Diag, for diagnosis versities, government, and industrial proceeds by repetitive
applications and based on ES/KERR&D centers. I was able to accom
sequences of passing messages NEL, and ES/Promote/W-Plan, for pany this group on many of their to a method within a frame, planning and scheduling applications visits. The following report is an
and method execution.
that are based on Whispa. (I was overview of one particular research • Assumption-based: Reasoning told that the next generation of a organization, Hitachi, and their spe- starts with a number of
domain scheduling shell will be based cific programs. I have organized this possible solutions and solutions on Tosca, and will incorporate extento first catalog some specific exam
that do not meet the specified sive user feedback.) Other packages ples of expert systems that have been conditions are discarded. This include a knowledge acquisition tool developed by and for Hitachi cus- can be used in the production ES/Tool/W-RI rule induction from tomers. Thus these illustrate some
system as a frame.
examples, a shell for real-time proexpert systems that are in use here in • Multilayered: This is a cess control, another for plant operaJapan. Next we describe major re
communication mechanism tion, and a user interface building search projects underway that related that executes reasoning
tool UIBT that has some multimedia to the topics of AI and ES and con- through cooperation between capability. There is also ES/Guide for nections to other research around
different expert systems.
Expert System building support the world.
based on the spiral and waterfall The main software tool provided Knowledge is represented by a model. Unit sales of these other by Hitachi is a version of ES/KER- combination of rules and facts – A tools are a small percentage of the NEL. (Separately named versions rule is essentially an IF (condition) ES/KERNEL sales. Some fuzzy exist for workstation, mainframes, THEN (execute something). Facts reasoning has been implemented in PCs, and for On-line processing. are either represented as frames or this package. There is also a new workstation ver- private memos. A frame is used to Hitachi has implemented Expert sion ES/KERNEL 2 that runs under
In ES/KER- Systems (mostly via ES/KERNEL)
represent static facts.
in a large variety of application from production conditions and raw Government
Oil transformer breakdown diagHitachi points out that "even though User (repairman) enters details of
nosis-User enters type and volume the range of application of expert breakdown, and system produces
of gas. (Gas contained in a sample
of insulation from transformer locatsystems is very wide, only advanced breakdown cause interactively. firms are engaging in their develop
ed in office or building is extracted ment, and then in limited areas."
and analyzed.) Inference is perConstruction
formed to pinpoint cause of trouble Thus, like the United States, smaller companies have not yet embraced
and next inspection date is deter
Shield tunnel planning support- mined. this technology. The first list below
User enters construction period, Landslide diagnosis-Power gencontains material given to us by
diameter of bore, route, soil quality, Hitachi in very abbreviated form.
eration facility sites are checked to The second lists the titles and organi
volume of groundwater, and other see if they are prone to landslides.
conditions, and the system deter- Soil nematode diagnosis-Diagzation of expert systems in use that
mines ideal shape of cutter, number noses whether or not crop damage is were provided to us in much more
of cutters, and positioning of shield detail (omitted here). Thus, we see
caused by nematodes. If so, deterboring machine. This system required mines nematode type and displays that major application areas are in
about 60 person months to complete high volume large transaction pro
method to eliminate the problem. and a cost of about US$500K. It cessing, where the actual reasoning is
Building inspection Supports contains about 9K lines of C, 100 shallow. Finally, Hitachi pointed out
building inspection in areas of deframes, 850 rules, relates to about that currently the most important
sign, structure and facilities, to check 250 parts on a shield machine, and whether or not they conform to perapplication areas for Expert Systems
accesses a relational database of are for scheduling. Less important,
mits and applicable codes. about 15K items. but still significant, are for diag
Elderly welfare consultation
Foundation improvement method nostics. This is also the case in the
User enters personal data that then selection support—User enters the judges status and procedures. System United States. status of subsidence and the struc
also outputs nursing home entrance Manufacturing
ture to be built, and the system pro- eligibility.
duces optimum foundation improveBlast furnace power generation ment method.
Securities support-User enters information on the status of turbines or other equip
Investment portfolio creation ment, and system displays diagnostic Distribution
support—Sales support system to cremessages and results of inferencing.
ate investment portfolio for clients Melting (steel production) plan
from a large variety of products. ning support-Optimum stock of raw Work scheduling
Terminal configuration managematerials and most economical melt- management system for super-mar
ment-Intelligent terminal manageing plan is created according to re- kets. Performs inference based on
ment for online system. Automates quired monthly ingot production weather, local conditions and Point
setting of parameters when new tervolume. of Sale data.
minal is introduced. Drainage system constructionFacing (merchandise distribution)
Stock investment consultationUser enters size of field and quality control-Creates merchandise dis
Technical analysis of a large number of soil, and the system produces de- tribution (display) and configuration
of indices and the experience of sign drawings and estimate of ex- plans. pense and time required to complete Dispatch plans-Predicts delivery
stock experts is used to conduct stock
volume for the day from past data Resin production planning sup- and also displays required number of Insurance port User enters sales data and trucks for each area. furnace operating status, and system
TV channel selection budgeting
Claims assessment Judges produces integrated production plan Plans for TV commercials are creat
whether or not claim made by client from sales to detailed production ed and budgeting is performed for
for hospitalization, etc., should be process. Knowledge base formed specific regions.