Journal of the Japanese Society for Artificial Intelligence
Online ISSN : 2435-8614
Print ISSN : 2188-2266
Volume 5, Issue 5
Displaying 1-36 of 36 articles from this issue
Print ISSN:0912-8085 until 2013
  • [in Japanese]
    Article type: Preface
    1990 Volume 5 Issue 5 Pages 525
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (63K)
  • [in Japanese]
    Article type: Corner article
    1990 Volume 5 Issue 5 Pages 526
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (110K)
  • Noriaki SANECHIKA
    Article type: Corner article
    1990 Volume 5 Issue 5 Pages 527-537
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (1031K)
  • Masayasu ATSUMI
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 538-547
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    Recently, researches for knowledge system development environment which supports advanced knowledge representations and reasonings have been progressing. The purpose of the WORLDS research is to develop one of these environment, which equips the multiple viewed-worlds mechanism for an assumption-based reasoning in incomplete knowledge environment. This paper presents the design concept and implementation of the knowledge system/language WORLDS and its simple application. The main features of WORLDS are summarized in following three points. 1) It has a focus of attention control mechanism, which attends to a subset of rules and a subset of objects. This mechanism constructs a hierarchical working memory between viewed-worlds and a knowledge base or an assumption set by adding premises from a knowledge base to viewed-worlds and generating assumptions from an assumption set successively. 2) It has a multiple viewed-worlds mechanism, which extends viewed-worlds based on a focus of attention control, monitors inconsistency among viewed-world nodes based on nogood relations and maintains viewed-worlds based on truth maintenance. 3) It has an amalgamated knowledge representation and inference method of object-oriented, rule-oriented and logic-oriented forms, which is planned to be suitable for an assumption-based reasoning by multiple viewed-worlds mechanism. This amalgamation is accomplished by logical interpretation of message passing and rule execution based on message passing. A simple knowledge base system for city planning support has been built using WORLDS's knowledge representation and reasoning facilities, and its ability has been confirmed.

    Download PDF (983K)
  • Takeshi CHUSHO
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 548-557
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    One major defect in neural network models in earlier research is nonclarity of computational Semantics. This paper describes a neural network system for thoutht process simulation called "Repetition of Association and Concentration Model". Repetition of concept association is simulated by a neural network model with a feedback loop. However such a loop results in a stable state which is contradictory to the dynamics of the thought process. In this model, dynamics are maintained by nonlinearity to simulate consciousness, that is, thought energy distribution is enhanced to facilitate concept recollection. Computational semantics of this model are defined as a parallel production system by correspondence of the synaptic connections between neurons to production rules. Three types of learning facilities are described as automatic knowledge acquisition. The relevance of this model is suggested by simulating the learning procedures for recognition of relations between concepts.

    Download PDF (888K)
  • Yasuhiro KOBAYASHI, Motoaki UTAMURA, Takao SATO, Daijirou KATAYANAGI, ...
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 558-568
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    A planning system PREMIS has been developed as a tool for preventive maintenance of plant components. Input data of PREMIS includes plant name, inspection frequency and observed events in operating period prior to inspection. From maintenance engineer's expertise in knowledge base and function test data from component data base, PREMIS selects components to be inspected at specified frequency with justification of selection, and detailed explanation if necessary. The system is characterized as follows : (1) Use of generic and flexible frame work of knowledge base configuration and inference control for problem-solving in planning tasks by monitoring-based and scheduling-based preventive maintenance methods. (2) Integration of knowledge base and data base realized due to requirements in preventive maintenance domain. (3) Use of man-machine interface with datailed explanation facility. On the basis of PREMIS, a component-specific system PREMIS-CRD has been built for control rod drive (CRD) mechanisms in nuclear power plants. PREMIS-CRD has been applied to a case of selection of 31 out of total 185 CRD's. It was confirmed that the result agreed to that by an expert planner. Machine time for this case was 2 min. use and 10 sec. cpu on a workstation HITAC-2050.

    Download PDF (941K)
  • Hideki ISOZAKI
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 569-577
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    Current qualitative reasoning systems have many problems. One problem is complexity of computation. As qualitative simulation is based on combinations of qualitative values of physical quantities and their constraints, notorious combinatorial explosion occurs easily. Another problem is lack of guidelines for representing and reasoning about spatial curves or surfaces. Such guidelines must be useful when we simulate human reasoning by computers. In this paper, we propose a new qualitative representation of curves to solve these problems. This representation is called "Zero Point Lists (ZPL)". Each ZPL is an ordered list of zero points of several functions (qualitative variables) along an independent parameter s. Often ZPL is extended to keep more information. When signs of two variables are assumed to be equal or opposite in some intervel of s, ZPL is extended to keep the relationships. Though it is straightforward to translate from ZPL notation into conventional sign notation, the former representation requires less branching, because each ZPL corresponds to several qualitative behaviors represented in ordinary sign convention. We define fundamental ZPL operations such as differentiation, addition and multiplication. These operations are useful to estimate qualitative behaviors of a system after observing it several times. These rules are also used to infer behaviors from constraints on the qualitative values of physical quantities.

    Download PDF (788K)
  • Kouji IWANUMA, Masateru HARAO, Shoichi NOGUCHI
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 578-587
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    In this paper, we give a new method for computing parallel circumscription based on an equivalent transformation of queries. Circ [A;Γ;Δ] ⊧ B, is equivalent to A ⊧ B if B has no negative occurrences of predicates of Γ and no occurrences of predicates of Δ. Therefore, if B satisfies these conditions, then the computation of Circ [A;Γ;Δ] ⊧ B can be reduced into the first-order computation of A ⊧ B. In general, the queries B don't satisfy the above conditions, but, sometimes, they can be transformed into the above computational sentences. In this paper, we present an equivalent transformation method of queries into the above computational sentences, and formalize it in the form of the resolution. This transformation method replaces a negative occurrence of a minimized predicate with the formula representing its minimal extension. It is similar to the unfolding in logic programming. This transformation method is intuitionally very clear, and can deal with theories consisting of both recursive and non-definite clauses.

    Download PDF (841K)
  • Hideyuki NAKASHIMA
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 588-594
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    It is now commonly accepted that we have to refer to the surrounding situation to make proper inference. We will further consider inference which treats situations as objects. We will also apply this idea to Yale shooting problem, and show that it gives us a framework to represent the problem reflecting human tendency toward a particular solution according to the situation. we will borrow the basic framework from situation theory which treats information with respect to situations. In the theory, the truth value of a formula cannot be determined without actually referring to the situation which supports the formula. We Will consider inference whose rules, as well as formulae, depends on the situation. The inference must be capable of talking about the situation in which it is carried out. We call this "inference about the situation" to contrast with "inference in the situation".

    Download PDF (660K)
  • Masumi ISHIKAWA
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 595-603
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    Learning in connectionist models has two aspects : the first aspect being the reproduction of the mapping from input to output patterns, and the second being the discovery of regularity in these training patterns. The backpropagation learning algorithm stresses the former aspect, as can be seen from its criterion function of the sum of squared output errors. The present paper, on the other hand, lays emphasis on the latter aspect. In the backpropagation learning of feedforward type models it is of nesessity to determine, beforehand, the number of layers and the number of hidden units in each layer. Since this prior determination is, in general, difficult, a trial and error procedure is inevitable, which is quite time consuming. To overcome this difficulty and to generate a small sized network, the present paper proposes a learning algorithm with forgetting of link weights. This forgetting is realized by adding the sum of absolute values of link weights to the criterion in the backpropagation algorithm. This algorithm generates a skeletal structure, in which the numbers of links and units used are kept as small as possible. As by-products of this algorithm it has various advantages : ease of interpretation of hidden units and improved generalization power of the resulting models. This algorithm alone causes the following two difficulties : emergence of distributed representation on hidden layers, which makes the interpretation of hidden units difficult, and a poor criterion value after learning due to the added term in the criterion function. To resolve these difficulties a structural learning algorithm is proposed, which consists of a series of algorithms : the learning algorithm with forgetting, a hidden units clarification algorithm, and a learning algorithm with selective forgetting. This structural learning algorithm is applied to a problem of discovering a logical function from a given set of pairs of input and output logical values. It is well demonstrated that the resulting skeletal network represents logical structure of the given problem. On the contrary the backpropagation algorithm generates a network far from being skeletal, making the interpretation of hidden units quite difficult. This algorithm is applied to another problem of classifying iris data by Fischer.Generalization power of the structural learning algorithm and that of the backpropagation algorithm are compared. The result of the comparison clearly demonstrates that the former has greater generalization power than the latter.

    Download PDF (815K)
  • Yasumasa OOMORI, Haruki UENO
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 604-616
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    Most currently available knowledge-based expert systems based on a surface knowledge, such as a production rules, can work only for the specific problem solving for a specific problem in a narrow domain. This is because that one rule allows single interpretation and then a complex problem needs a lot of rules according to a variety of situations. For this reason it is difficult to achieve flexible and powerful system by this kind of knowledge. While, a human expert who has the deep knowledge according to the domain can solve variety of problems in the domain using the knowledge. By deep knowledge we mean the knowledge about the principle and the structure on an object. For example, a human expert who knows the principle and the structure of a computer system can express what the computer is to another person, can understand a picture about the computer,can find some failures in the computer design specification, and can make some troubleshootings on the computer. We believe that features to manipulate the deep knowledge might become key technology for next generation expert systems. We have proposed the concept and the modelling method based on the object model as a candidate of deep knowledge system. We developed a trouble-shooting expert system as an application of the object model. The object model can be represented in a extend frame-based knowledge representation formalism by means of a combination of ISA and PART-OF relation, and is represented by combining knowledge about structures and behaviors, and is one of deep knowledge for representing complex structured objects. The behavioral knowledge is represented the relationship between input and output states. We called that behavior rules. The model-based reasoning method using the behavioral knowledge is achieved by evaluation of behavior rules. This paper shows the object modelling and the model-based reasoning based on the object model. The system have demonstrated the troubleshooting of a focal plane shutter. The system was developed using the frame-based knowledge engineering environment ZERO on the AI work station ELIS.

    Download PDF (1122K)
  • Kouji IWANUMA, Masateru HARAO, Shoichi NOGUCHI
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 617-626
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    The computation of Circ [A ; Γ] ⊧ B can be reduced into the first-order one of A ⊧ C, if the query B can be transformed into a sentence C such that C is equivalent to B and has no negative occurrences of any minimized predicates p∈Γ. In the previous paper, we presented a rule for equivalently transforming the queries B into the above computational sentences C. But, in general, by this transformation rule, a query may be transformed in several ways. Therefore, it is important to consider a normalizing strategy, which controls the query transformation in such a way that the above computational sentence, i. e., a normalized sentence, can be always obtained if such a sentence exists. In this paper, we investigate normalizing strategies for the above equivalent transformation of queries in predicate circumscription. As a result, we show that a fair transformation strategy is a normalizing one. Through the proof, we also show that the query normalization can be regarded as a kind of the derivation of a finitely failed SLD-tree in logic programming.

    Download PDF (905K)
  • Kazuhisa MIWA, Noboru SUGIE, Moriya ODA
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 627-639
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    In this paper, we discuss inductive inference which is a kind of learning. We consider this problem through "a guessing task of a two-dimensional arrangement of cards". This task is summarized as follows. Four kinds of color cards are arranged regularly at a 4 × 8 two-dimensional domain. All cards are turned-off at an initial state. From the initial state, a subject (learner) tries to find a rule of the card arrangement under a constraint of minimum number of turning-up cards. Firstly, we describe regularity of patterns of the card arrangement by theory of groups. In this description method, regularity of patterns is defined by "unit structure" and "repetition structure". We indicate efficiency of the description method by theory of groups through analysis of regularity of patterns generated by human subjects. Secondly, we construct an algorithm which can solve the guessing task on the basis of the description method by theory of groups, and realize the algorithm on a computer. Results of computer simulations are summarized as follows. Inference by this algorithm prescribes the upper boundry of inference by human subjects. This algorithm could guess correctly 95% of patterns generated by human subjects.

    Download PDF (1118K)
  • Toshiro MAKINO, Mitsuru ISHIZUKA
    Article type: Technical paper
    1990 Volume 5 Issue 5 Pages 640-648
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS

    The hypothetical reasoning, which is one of non-monotonic reasoning schemes, is a method for handling incomplete knowledge in knowledge base. It also provides us a practical framework for building diagnosis and design systems. In this paper, we present a hypothetical reasoning system with constraint handling mechanism which is indispensable for pruning search space in design problems. Then, we describe a circuit-block design system constructed on this framework. Although the concept of applying this reasoning system to design problems has been reported previously, there has been no its practical application to design problem. In response to the specification of behavior and some constraints given to the system, our system can synthesize a circuit which satisfies the specification and constraints by combining function blocks given in advance as design components. This system differs from existing rule-based design expert systems, since the system builder of this system is only required to write down basic knowledge and doesn't need to write heuristic knowledge for guiding the inference process. Instead of using such heuristic knowledge,a constraint-guided inference process implemented in the hypothetical reasoning becomes the main power for synthesizing new design patterns. Thus the knowledge acquisition burden is greatly lessen and the scope of competence of the knowledge-base can be broaden by allowing incomplete knowledge in the knowledge-base. This paper shows how a practical design system can be effectively implemented using the framework of the hypothetical reasoning system.

    Download PDF (831K)
  • [in Japanese]
    Article type: Other
    1990 Volume 5 Issue 5 Pages 649
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (104K)
  • [in Japanese]
    Article type: Corner article
    1990 Volume 5 Issue 5 Pages 650-652
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (283K)
  • [in Japanese]
    Article type: Corner article
    1990 Volume 5 Issue 5 Pages 653-654
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (149K)
  • [in Japanese]
    Article type: Corner article
    1990 Volume 5 Issue 5 Pages 655-656
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (217K)
  • [in Japanese]
    Article type: Corner article
    1990 Volume 5 Issue 5 Pages 657-658
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (113K)
  • Article type: Other
    1990 Volume 5 Issue 5 Pages 659
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (76K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 660-664
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (347K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 665-670
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (588K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 671-694
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (1837K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 695-697
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (186K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 698-703
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (323K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 704
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (51K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 705
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (32K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 706
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (61K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 707
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (51K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages 708-709
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (173K)
  • Article type: Activity report
    1990 Volume 5 Issue 5 Pages b001-b012
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (858K)
  • Article type: Other
    1990 Volume 5 Issue 5 Pages b013-b015
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (171K)
  • Article type: Other
    1990 Volume 5 Issue 5 Pages b016-b022
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (419K)
  • Article type: Cover page
    1990 Volume 5 Issue 5 Pages c005
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (28K)
  • Article type: Cover page
    1990 Volume 5 Issue 5 Pages c005_2
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (86K)
  • Article type: Table of contents
    1990 Volume 5 Issue 5 Pages i005
    Published: September 01, 1990
    Released on J-STAGE: September 29, 2020
    MAGAZINE FREE ACCESS
    Download PDF (71K)
feedback
Top