In this paper, we propose a method for generating macro-models of mobile multi-agent systems to explain their emergent cooperative behaviors at macro-level. As an example for this method, we choose foraging behaviors of ant colony, which is a typical natural multi-agent system and shows complex cooperative behaviors without any central control nor higher cognitive mechanisms. In the method, a macro-model generator receives macro-behavior rules which are descriptions of the condition and the effect of phenomena observed in a target system, and the reasoner generates a macro-model which includes physical phenomenon models, qualitative regions(description of spatial extents), their geometric relations, and causal graphs consisting of monotonic and differential relations among state variables. We report the results of numerical simulations for a simple model case of foraging and the way of generating their macro-models.

Following the simple model of ant colony's foraging, we introduce a biological supposition called desensitization mechanism, which makes an ant lose its sensitivity to pheromone signals temporarily, in order to avoid over-concentration to a single bait-site observed in the simple model case and to realize decentralized control. We examine the validity of the method for macro-model generation, by the comparison between our method and the numerical simulation results, where causal graphs generated by our method can explain the global and long-term tendency of macro-behaviors of the target system on mathematical basis, for both a simple model case and a decentralized control with desensitization supposition.

In this paper, we describe a GA system based on variable length chromosomes. Our goal is to establish "analog evolvable hardware", i. e., analog EHW.Analog hardware is vulnerable to the environment, such as temperature, and traditional methods to deal with the environmental disturbance are complicated and ad-hoc so that they are not fit for designing a large-scale circuit. In the EHW approach, noises and errors are fed back from the manufacturing process to the designing process. These processes are repeated until the adapted hardware is acquired. Thus, the EHW is robust to the external disturbance. This paper introduces a new approach to the analog EHW. We use a component-list representation, which has the advantage of enabling the design flexibility, i. e., the capability of acquiring more various functions. We can also properly size a circuit without designer's specification. Our system also features the two-stage evolution, i. e., the structural adaptation of the circuit topology and the parameter adaptation by tuning up the elements'values. This division is aimed at achieving more efficient global search with local modification. We show experiments which support the superiority of our approach over conventional approaches in terms of robustness and computational cost.

We investigated collaborative discovery processes using interactive production systems. we controlled the degree of sharing cognitive space:(a)the experimental space was only shared, (b)the hypothesis space was also shared, and(c)these two kinds of space were integrated. We compared the performance of finding targets in the three types of collaboration above with the performance in the case of two systems independently finding targets. If the performance in the former cases exceeds that in the latter case, we approve of emergence brought by interaction by the two systems. The results of the computer simulations were consistent with our ordinary knowledge on collaboration. That is, as the extent of sharing common space is larger and two interactive systems perform deeper collaboration, the effects of interaction gradually increase. If two systems obtain the maximum effects of collaboration by exchanging intellectually mutual information, collaboration produces emergence.

This paper presents a method of generating negative instances based on similarity between positive instances for inductive learning systems, which construct the concept descriptions from positive/negative instances of a target concept. Given only positive instances, such systems may face a problem of overgeneralization of the resulting descriptions. This problem prevents existing useful inductive learning systems from being applied to KDD(Knowledge Discovery in Database), since most databases consist of only positive instances. To solve this problem, We generate negative instances from positives instances. We concentrate on similarity between positive instances, and construct the "approximated space" of the target concept, which contains all positive instances and instances close to them on the basis of the similarity. Then the instances which are not in the approximated space but are most similar to the positive instances are generated as the negative instances of the target concept. Such negative instances behave as near-miss instances and protect the over-generalization.

Most of large vocabulary continuous speech recognition system employ speaker adaptation or speaker selection method. Many speaker adaptation and speaker selection methods require distance between speakers. However, there are many variations in definition of distance between speakers, and their performances are not investigated. In this paper, we propose a new distance calculation method for distance between speakers based on a structure of Hidden Markov Network(HMnet)made by SSS-free. A structure of HMnet represents speaker distribution because SSS-free constructs an HMnet based on acoustical feature of training samples. From the experimental results, the new distance showed better correspondence with speaker distribution than traditional distances.

Content-based retrieval of multimedia information has been investigated in several research projects. In this paper, we will focus on the automatic indexing method of motion information. Automatic indexing is performed in a pattern matching approach. Reference patterns are necessary for pattern matching, so that we will propose two methods to define primitive motions as reference patterns. One method divides motion data into segmental motion data by detecting the change of motion speed. The other classifies segmental motions such that similar segmental motions are gathered in the same cluster by nearest neighbor algorithm. In order to evaluate the similarity between two segmental motions, we use the dynamic time warping(DTW)method because each segmental motion takes different time length even if one person performed the same motions. Motion data can be converted into symbol sequences which represent primitive motions. Then continuous dynamic programming(CDP)method is used to recognize contents of motion. CDP is one of the extensions of DTW. It makes us possible to recognize a motion even if it is complex. At last, we use a bagging approach to merge the two similar clusters which have same primitive motions.

In this paper we propose a probabilistic model and an expert system which constructs image processing procedures to detect specific figures automatically based on the model. An image processing procedure is constructed when user inputs a set of pictures consisting of normal samples and abnomal samples with sketches corresponding to abnormal areas and also inputs requirements on misclassification rate. We consider the image processing prosedure as a kind of description of knowledge which a user acquires from experience. We also show experimental results of applying the expert system to a real world problem.

In Multi-Agent Systems, agents sense their local information, and they estimate the global environment. Autonomous agents should have independent estimate functions according to their local information. The functions should adapt the global environment. We consider this estimate function to be exactly a "norms of behavior" in the human sense. In this study, we construct a competitive social system with selfish autonomous agents. Agents interact with their alternative actions each other repeatedly as like as IPD(Iterated Prisoner's Dilemma)games.Agents have a estimation table for their action independently. Agents'strategies of action are adapted to their own estimate function, and agents'norms of behavior satisfy. But global environment deprives their energy according to a global rule. In such systems, the agent which adapt not only strategy but also norms of behavior to the environment can survive.

This paper introduces an automatically reasoning system for inferring curved-surface solids from inconsistent three views. Our system is based on a so-called knowledge-based method that imitates human reasoning process in inferring curved-surface solids from inconsistent views. Initially, curved-surface primitives are recognized and extracted one by one according to their possibility value computed from all the given three views. Next, those data only relative to curved-surface primitives are removed from three views, and then polyhedrons will be constructed. The final solid is formed by assembling all those generated curved-surface primitives and polyhedrons. Our system has been implemented in C and OpenGL on Windows NT. The constructed solid can be observed at free angles through rotating it with keyboard. Many examples are tested and the experimental results show that our system is very robust, and can accurately and effectively construct a suitable curved-surface solid from inconsistent three views.

Internet and agent technologies have facilitated world-wide trade, but we sometimes encounter risky situation in exchange processes involving goods and money, i. e., fraud. This problem is becoming more serious with the growing popularity of person-to-person trade. One of the reasons for such fraud is that obtaining a new identifier in netwoeks is cheap. This makes it difficult to exclude malicious agents from trade. One solution is the imposition of an entry fee. However, if an entry fee is expensive, it discourages newcomers from starting deals. To resolve the conflict between safety and convenience, we have developed two kinds of exchange mechanisms that guarantee against defection from a contract. One reduces the entry fee by integrating multiple deals and controlling goods and money flows. We also show an extension of this mechanism that can reduce the entry fee more while taking a longer time to complete the exchange process. The other reduces the entry fee by incorporating a third party agent into the exchange process. We examine the lower bound of the entry fee for each mechanism and provide a calculation method that is able to obtain this value in linear time.