Shun'ichi Tano

Interval-valued fuzzy backward reasoning

The importance and efficiency of backward reasoning in nonfuzzy reasoning has been stressed for a long time, especially in the case of expert systems and decision-support systems. The extension of this reasoning method to fuzzy theory, however, has never been considered. In this paper, the authors propose a definition of fuzzy backward reasoning based on the generalized modus ponens and show the necessity of considering interval-valued fuzzy backward reasoning. Then, the authors propose solving methods for fuzzy backward reasoning in the case of a rule with one or several conditions as well as in the case of several rules.

Deep combination of fuzzy inference and neural network in fuzzy inference software—FINEST

Abstract At the Laboratory for International Fuzzy Engineering Research in Japan (LIFE), we are now developing FINEST (Fuzzy Inference Environment Software with Tuning). The special features are (1) improved generalized modus ponens, (2) mechanism which can tune the inference method as well as fuzzy predicates and (3) software environment for debugging and tuning. In this paper, we give an outline of the software, and describe an important concept, a deep combination of the fuzzy inference and the neural network in FINEST, which enables FINEST to tune the inference method itself. FINEST is now being used as a tool for quantification of the meaning of natural language sentences as well as a tool for fuzzy modelling and fuzzy control.

Operator tuning in fuzzy production rules using neural networks

In production rules, the total matching degree of the condition part is calculated from the matching degree of each condition by aggregation operators. In ordinal production systems, simple logical AND and OR functions are used as aggregation operators because the matching degrees are crisp values. In the case of fuzzy production rules, there are several promising approaches for handling the uncertainties of matching degrees. Investigations have been conducted on the automatic tuning of membership functions using neural networks. If a complex relationship exists between the conditions, tuning methods are not able to adjust for errors by tuning the membership functions. If similar characteristics exist in the relationships in some rule blocks, it must be possible to efficiently tune the rule blocks by tuning the aggregation operators. From this standpoint, as one step, the authors study tuning the aggregation operators. They consider automatic operator tuning of parametric T-norms and T-conorms whose characteristics can be modified parametrically.<<ETX>>

Backward-chaining with fuzzy 'if. . . then. . .' rules

The authors consider the issue of backward-chaining when using fuzzy if-then rules. Usually, forward-chaining is used to induce new information, given a set of rules and a set of data. However, backward-chaining is more efficient when one wants to check whether a given fact holds. A mathematical formulation of the problem in the case of finite and continuous universes of discourse is given, and a method is outlined to solve the problem. An interpretation is given of the different possible solutions, and the problem of the choice of a particular solution is considered. As the method is very sensitive to the given goal, in the case in which no solution can be found the extent to which it is possible to modify the initial problem to find an approximate solution is examined. Some results relative to finite sets are generalized to the case of continuous sets.<<ETX>>

Conjunction and disjunction with synergistic effect

The authors present a method for forming combination operators that can account for synergism between the items that make up the conditions of a rule. When the condition part of a rule contains several conditions, the global truth value of the rule is given by the combination of the truth values of all the conditions. Usual combination operators, however, cannot be used to express synergistic relations that may exist between the different items in the conditions. Therefore a method for defining new combination operators that takes those relations into account is proposed.<<ETX>>

A rule-based method to calculate exactly the widest solution sets of a max-min fuzzy relational inequality

Abstract In this paper, we propose a rule-based method to calculate exactly all the widest solution sets of a fuzzy relational inequality. Viewed as an extension of fuzzy relational equations, fuzzy relational inequalities can be applied to many fields of fuzzy logic. Yet, they provide with more flexibility, as they enable to get rid of the sensitivity of fuzzy relational equations to numerical values. Up to now however, fuzzy relational inequalities have received few attention from researchers, and few methods have been proposed to solve these inequalities. The proposed analytical methods, furthermore, do not calculate exactly all the widest solutions sets of a fuzzy relational inequality, and often lead to redundant solution sets. The rule-based method we propose in this paper determines easily exactly all the widest solution sets of a fuzzy relational inequality.

New Design Concepts for the FLINS-Fuzzy Lingual System: Text-based and Fuzzy-centered Architectures

A fuzzy natural language communication system called the Fuzzy Lingual System (FLINS) is currently under development at the Laboratory for International Fuzzy Engineering Research (LIFE). The final goal of the FLINS project is to create a lingual computer, that is, a domain-independent teach, question, and answer (TQA) system. In this paper, we propose two new design concepts, text-based architecture and fuzzy-centered architecture, to realize our goal. Two experimental systems were built to determine the feasibility of those approaches before the development of FLINS. One is a text-based natural language understanding system, called the AB-System. The other is a fuzzy expert system called FOREX, which predicts exchange rate trends according to fuzzy rules and fuzzy data.

Effectiveness of annotating by hand for non-alphabetical languages

Unlike documents, annotation for multimedia information needs to be input as text, not in the form of symbols such as underlines and circles. This is problematic with keyboard input for non-alphabetical languages, especially the East Asian languages such as Chinese and Japanese, because it is labor intensive and imposes a high cognitive load. This study provides a quantitative analysis of the effectiveness of making annotations by hand during a note-taking task in Japanese. Although the lessons learned from this study come from Japanese text input, they are also generally applicable to other East Asian Languages which use ideographic characters such as Chinese. In our study, we focused on both the ergonomic and cognitive aspects and found that during annotation and note-taking task input by hand is more effective than input by keyboard. Finally, we anatomized the keyboard input problem and discuss it in this paper.

Fuzzy natural language communication system-FLINS: concept and conversation examples

The Fuzzy Computing project (FC project) at the Laboratory for International Fuzzy Engineering Research (LIFE) in Japan is developing a natural language communication system called FLINS, which is short for Fuzzy Lingual System. The final goal of the project is to implement a lingual computer that can communicate and learn, both by being taught and on it own, through use of fuzzy natural language. This paper gives an overview of the FC project, especially FLINS. It first briefly presents the key concepts of FLINS, then explains the three-layered fuzzy inference and the inference control based on the three-layered inference that are the special features of FLINS, and finally gives conversation examples.<<ETX>>

Fuzziness Reduction Method for a Combination Function

Conventional fuzzy reasoning has a well known problem that the fuzziness of inferred results gradually increases according to the progress of the inference. Essential problems are (PI) the membership value of the combined result never approaches to grade 0, but always increases, and (P2) lack of reinforcement property. We proposed a new combination function which resolves the problems by introducing equilibrium E and dependency factors α and β represented by stochastic rules. Its behavior is consistent with that of a human. Moreover, it covers the range of conventional combination functions.