Production systems are established methods for encoding knowledge in expert systems. In production systems, it's an important problem how to deal with expensive productions. Expensive productions are rules which would be required the extraordinary time and space to match (for each rule, compare the left-hand side against the current working memory). We propose a method to change two match algorithms, RETE and Direct-match algorithm, dynamically during execution.

Many problems in multi-agent systems can be described as distributed Constraint Satisfaction Problems (distributed CSPs), where the goal is to find a set of assignments to variables that satisfies all constraints among agents. However, when real-life problems are formalized as distributed CSPs, they are often over-constrained and have no solution that satisfies all constraints. This paper provides the Distributed Partial Constraint Satisfaction Problem (DPCSP) as a new framework for dealing with over-constrained situations. We also present new algorithms for solving Distributed Maxima1 Constraint Satisfaction Problems (DMCSPs), which belong to an important class of DPCSP. These algorithms are called the Synchronous Branch and Bound (SBB) and the Iterative Distributed Breakout (IDB). Both algorithms were tested on hard classes of over-constrained random binary distributed CSPs. The results can be summarized as SBB is preferable when we are mainly concerned with the optimality of a solution, while IDB is preferable when we want to get a nearly optimal solution quickly.

Deduction by a computer studied so far has been centered around symbolic reasoning with formulas. Recently, attention has been directed to reasoning with diagrams as well, in order to augment the deficiency of reasoning with symbols only. In this paper, we consider the deductive reasoning method with diagrams (Venn diagram) and its proof complexity. For this, we examined the three aspects of differences between symbolic reasoning and diagrammatic reasoning: proof complexity and proof similarity, by proving the validity of a variety of Aristotelian syllogisms both symbolically and diagrammatically. The results we have found include that the proof steps by Venn diagram are much shorter than those by natural deduction, and there are eight different diagrammatic proofs and only two different diagrammatic proof patterns for fifteen valid Aristotelian syllogisms.

Recently, multi-agent systems have been studied diversely. The authors have proposed the iterated continuous prisoner's dilemma game in order to study on elastic interactions between agents. This is a continuous version of the iterated prisoner's dilemma game. Each agent does not choose his action from two alternatives but from a continuous range in this game. So far, one-to-one plays have been analyzed on the invasion relation. In this research, we extend this relation to aprocess. Namely, we define the process of invasion as three phases, 1) invasion of a single strategy, 2) invasion by a cluster, and 3) invasion by a group. The purpose of this research is to investigate the profitability of a strategy taking an intermediate action in the process of invasion and to clarify differences between this profitability and that of a strategy choosing his action from two alternatives. TFT is a well-known good strategy in the iterated prisoner's dilemma game. A group of TFT's is not invaded in the case that a strategy chooses his action between two alternatives. We also focus on TFT in this research. TFT can take intermediate actions in our continuous version. We analyze the process of invasion upon a group of TFT-like strategy DTFT's taking only discrete actions. Then we compare the profitability of DTFT with that of TFT. And in order to investigate taking intermediate actions explicitly, we consider another TFT-like strategy TFT^g. We define TFT^g modified TFT so as to take inexact actions, where we call g an inexactlizing function. Namely, the inexactlization of actions means changing original actions to more intermediate values. Then we analyze the process of invasion by TFT upon a group of TFT^g's and so on.

Autonomous state abstraction, i.e., the process to acquire the general information from the primitive sensor inputs is a key issue in the research area of intelligent autonomous agents, because the sensor spaces for these agents are generally vast and it is almost impossible to treat each sensor state individually. At the same time, fusion of heterogeneous and redundant sensor information is indispensable for those agents to deal with the uncertainty in the real world. This paper proposes a new abstraction method based on Bayesian classifier, which enables the agents to integrate heterogeneous and redundant sensor information in a flexible manner. Simulation results have shown that this method increases the robustness of the agents' behavior against the disturbance such as sensor noise and fault.

In the research area of ILP (Inductive Logic Programming), the generation of new intermediate predicates is known to be a hard problem, which requires very large search spaces. In this paper we define the class of regular logic programs, having similar properties to the class of regular languages, and show an efficient algorithm which infers them using membership and equivalence queries. In particular, without requiring examples of intermediate predicates it can invent, such predicates dynamically and derive their definitions. The algorithm is based on the use of program schemata-ordered lists of the arguments appearing in the program clauses representing the target regular tree relation. Using program schemata we can extend Angluin's algorithm for inference of regular languages so that we can apply it to the inference of regular logic programs. The algorithm is realized with a routine for revising predicate decision trees and a new implementation of contradiction backtracking. The total running time of our algorithm is bounded to polynomial in the number of clauses m of the program presenting the target relation, and the maximum size w of user-provided counter examples.

Although GAs are widely used for optimization problems and have obtained many good results, these are also holding problems, such as premature convergence problems. In this paper, the premature convergence problems caused by reduction of the diversity in GAs are observed, and a new genetic algorithm, named multi-step GA (MSGA) is proposed. With MSGA, the diversity can be recovered for every certain segment of generations with keeping the evolution result till then. MSGA is the algorithm using hill climbing method as frame work, which starts some phenotype as an initial value and changes the value to the better direction. To evaluate proposed MSGA, function optimization problems and shape design problems are applied. As a result, MSGA can avoid the premature convergence and shows higher performance than the other conventional GAs for some problems.

Although many linguistic and psychological studies have been made on verbal irony, they still cannot distinguish ironic utterances from nonironic ones completely. This paper argues that irony is distinguished from nonirony in accordance with both the ironic environment condition (i.e., irony presupposes a situational setting consisting of the speaker's expectation, incongruity between the expectation and the reality, and the speaker's negative attitude) and the implicit display condition (i.e., irony satisfies at least two of the three features: allusion, pragmatic insincerity and indirect cues). This paper then shows that the argument is empirically supported, especially that ironic environment is a crucial distinctive feature of irony while implicit display suggests a high possibility of irony. It also suggests an interpretation process of irony consistent with these findings.

In this paper we propose a new ILP method Bottom Reduction. It is an extension of Bottom Generalization for treating multiple examples. Using Bottom Reduction we can find all hypotheses which subsume all of given examples relative to a background theory in Plotkin's sense. We do not assume any restriction on the head predicates of the examples. Applying Bottom Reduction to examples which have a predicate different from those of other examples, we can reduce the search space of hypotheses. We have already implemented a simple learning system named BORDA based on Bottom Reduction on a Prolog system, and we present, in this paper, some techniques in its implementation.

We propose a new model of sound stream segregation by using amplitude co-modulation and frequency co-modulation as auditory cues. This model adopts a hierarchical data structure which has not been used explicitly in the previous models of sound stream segregation. As a result, the hierarchies of streams were naturally represented in this model. In addition, this model is considered to have a potential of real-time processing and to give easy handling of adding constraints on sounds to be segregeted owing to modularity than previous modds did.

A logic-based hypothetical reasoning is an effective system that enables to apply to the problems such as a design and diagnosis, by handling incomplete knowledge as a hypothesis. However, inference speed has been slow, and that has been serious problem. There is a proof that hypothetical reasoning is NP-complete even on propositional logic. Therefore, many high speed hypothetical reasoning methods have been proposed. There are hypothetical reasoning systems that require all solutions, such as KICK-SHOTGAN or KICK-HOPE. KICK-SHOTGAN is a fast hypothetical reasoning system on propositional logic with inference-path network. KICK-HOPE is a fast hypothetical reasoning system on predicate logic and it is anapplication of QSQR method in deductive data base technology. In this paper, we carried out a comparative experiment of running time for these 2 systems, KICK-SHOTGAN and KICK-HOPE. First, we use knowledge-base written in propositional logic, and compare running time of these 2 systems. And next, we use knowledge-base written in first-order predicate logic. From consequence of this experiment, we clarify superiority or inferiority of both systems.