For verification, synthesis and design of hardware systems, various formal specification languages have been used in a combined form, which are temporal logics, finite state machine languages, data flow graphs and hardware description languages. Furthermore, informal natural language expressions are supplemented for the unified descriptions of time, action, state transition and causality, and for the abstract descriptions such as functions and hierarchy of devices. This paper proposes a method for situational and dynamic interpretation of natural language descriptions as well as formal language ones in hardware design specifications. Natural language expressions are formulated by the multi-modal predicate logic integrating logics of time, action, causality, conditional and knowledge. The semantics of the logic is given by first-order logic formalization based on verification-conditional and situational semantics. The satisfiability of logical formulas is defined by the provability in a situation description, which consists of first-order logical formulas representing factual situations, constraints and general rules in hardware systems. This paper presents the methods for transformation of natural language expressions into the first-order logical formulas through the multi-modal logical ones, and for description of the rules on modality, device action, register transfer and control state transition by the first-order logic. An efficient method for dynamic interpretation using belief revision is given which is based on the inference mechanism of the first-order logic and is reinforced with default reasoning and meta-level reasoning. The above processing system is implemented by Prolog language.

A hypothetical reasoning is a useful knowledge-processing framework applicable to system diagnosis, design, etc. Hewever, due to its non-monotonic inference nature, it takes exponential computation time to find a solution hypothesis set to prove a given goal. As for its propositional-logic version, we have shown previously that a polynomial-time approximate solution method of O-1 integer programming, Pivot and Complement method, can be effectively employed by transforming the hypothetical reasoning problem into a O-1 integer programming. Also, a networked-bubble-propagation (NBP) method constructed by reforming this approximate method into a networked mechanism has been presented to achieve an even faster inference. In this paper, we extend the NBP method to deal with predicate-logic hypothetical reasoning with variable expression but function-free. Furthermore, the network used for the inference (short term memory in this paper) stays within a reasonable size, since it is extended in a stepwise manner iff the current network has insufficient element hypotheses and rules to prove the given goal. With this extended NBP method, it is shown that a polynomial-time hypothetical reasoning can be achieved against the number of possible element hypotheses even in the case of predicate-logic knowledge.

An inference method for efficient processing of ordering constraints has been developed through combined use of look-ahead and constraint filtering in the generate-and-test like framework for constraint satisfaction, and applied to a knowledge system tool for job shop scheduling. The inference method is useful for providing a practical, tentative solution in scheduling problems where no feasible solution is available or where the existence of a feasible solution is unknown. A knowledge-based scheduling system with the proposed inference method has been generated and applied to job shop scheduling problems with 10, 100 and 200 jobs. From the results, it was concluded that the inferelnce method gives a tentative "quasi-feasible" solution and the optimum solution of job shop scheduling problems and that scheduling systems can be automatically generated owing to the precise formulation with assignment functions. It was confirmed that the constraint filtering contributes to the suppression of computer efforts of constraint satisfaction and that the look-ahead steps are optimized from computation trade-off between look-ahead and backtracking.

This paper describes a method to infer the initial values of design variables for specifications satisfaction problems in design. The problems are given with specification variables, intermediate variables, design variables, and equality and inequality constraints among the variables. If some of specifications and ranges of all design variables are given, it is necessary to infer appropriate initial vaiues of design variables and the rest of specifications to solve and satisfy constraints. In this case, if a good case near to the givens is known, the initial values can adopt the case. If problems, however, demand a new design or constraints are unfamiliar, it is difficult to infer the initial values. To solve this difficulty, a method to infer the initial values for the given specifications is needed. In the proposed method, constraints are represented in Constraint Logic Programming (CLP) and the initial values are repeatedly inferred from the given specifications by a neural network (NN) which learns the relations among values of design and specification variables by some values pairs of them calculated with CLP. The inference continues changing the worst one of the pairs with a better pair calculated by CLP using inferred values gotten with NN untill the initial values converge. This method is applied to a design of laser graphic device and the validity of the method is confirmed.

Conventionally, the development of a knowledge application system has often been accompanied with the constraction of its underlying knowledge-base system. As a result, such a tight coupling of an application system with its employed knowledge-base system makes it difficult to share and reuse a knowledge-base among different applications. However, large scale knowledge-base should be shared by a wide variety of applications. The efficient share and reuse of knowledge-base among different applications is currently one of the most important research issues in building knowledge systems. Thus, it is important to establish the notion of knowledge-base independence among applications and knowledge-base systems. In this paper, this important notion is first defined by three types of independence : maintenance independence, program independence, and algorithm independence. Then, a dynamic inheritance deduction mechanism is proposed to achieve knowledge-base independence in the environment of object-oriented programming languages. The mechanism acts as the interface between the application and its knowledge-base system : if a message invokes a method, a query for the knowledge-base is automatically generated. The result of reasoning in the knowledge-base dynamically decides the class hieraechy (i.e., the inheritance relation) of the application. By doing so, a knowledge-base can be effectively utilized by different application systems without knowing the detailed structure, which realizes knowledge-base independence. We implemented the dynamic inheritance deduction mechanism using the Objective-C language and the DOT developed in our previous work for deductive and object-oriented extended term representation. Furthermore, based on the proposed mechanism, we constructed the enviroment DOT-ADE (DOT application development environment) for supporting application development.

The job-shop scheduling problem (JSSP) is one of the most difficult problem to solve using various methods. We describe a GA approach which provides a better solution of JSSP in this paper. In GA, all individuals of population simultaneously search the best solution, and each individual is evaluated by its fitness in the environment. Originally, the evaluation is independent of genetic operations, such as crossover and mutation. So we introduce an interpreter which interprets the representation on a chromosome (genotype) and synthesizes a solution such as Gantt chart (phenotype). Then we can treat many complex constraints as procedures in the interpreter, and we solve a scheduling problem in a real world application. Along the line, We find that the solution based on the simple sequential coding is better than previous efforts using GA, and introducing "cluster averaging method (CAM)," which makes each cluster consist of same number of individuals, it improves the solution. The simple sequential coding represents an order that we allocate a work of a job on a Gantt chart. It means that the predecessor gene will determine the performance of succeccor genes. In other words, the head gene has the most important role in JSSP. Then we can introduce the schema that the only head gene of a chromosome has an fixed value and other genes have not-care values. Each individual is divided into those clusters by schema. Using the simple sequential coding with CAM, we can get better result in JSSP by experiment.

Similarity metrics is a central issue in automatic reasoning system, especially in case-based reasoning and analogical reasoning. This paper proposes a method of similarity metrics by focusing the similarity of features of problems which are represented by frame knowledge expressions. For retrieving the most similar previous case and adapting it in solving new problem, similarity metrics are needed to provide the retrieving criterion and the information of adaptation. In our method, we assign a degree of similarity as a retrieving criterion between the features of new problem and past one, and using fuzzy integral to calculate the degree of similarity by comparing target frame which presents the features of new problem with source frame which presents the features of past problem. While calculating the degree of similarity, we also make a comparative frame which includes the information of matching source frame to target frame for adaptation. Finally we show some simulating results.

A methodology to identify task/domain ontologies is proposed in this paper. According to the methodology, task/domain ontologies that are indispensable for building reusable knowledge base are identified. The methodology has two kinds of ontologies in the different abstraction levels. The first one is based on a vocabulary applicable in common to the entire target field. With this vocabulary, knowledge engineers can discuss with domain experts, and build a expertise model in collaboration with domain experts. The second one is based on a generic vocabulary. By defining the common vocabulary in terms of the generic vocabulary, reusable task model that is called a knowledge model can be built. We evaluate the effectiveness of this methodology by applying it into substation restoration problems.

We have implemented a weather prediction system, WINDOM, which uses memory-based reasoning (MBR). The observation data from Japan Meteorological Agency (JMA)'s networks, AMeDAS and the surface observation centers, are used directly as a case database. WINDOM runs on the parallel computer AP1000 and predicts the weather around Tokyo several hours ahead by matching the past data in the database with the present observational data. In MBR, the weighting methods for features and its metric definition are important to get high accuracy. In the experiments, four weighting methods based on conditional probability and three calculation methods of similarity are tested, and the best pair of methods is selected empirically. The relation between accuracy and parameters such as the area of the observational data, the quantity of data, and prediction hour are shown. By using 9 years of observational data in the middle and the west part of Japan to predict whether the weather of Tokyo 6 hours ahead would be rainy or not, an accuracy of 87.2% is achieved. The average accuracy of prediction in Kanto-area is slightly worse than that of JMA, but in some prefectures our results are comparable with their average results.