An approach to multi-agent world (MAW) description as well as a formal model of agent is presented. The model takes into consideration both "intellectual" and "energetic" aspects of agent's performance. Development of the presented concept may enable creation of methodology of design and implementation of decentralized intelligent systems.

Back

Reasoning within such domains as engineering, science, management, or medicine is traditionally based on formal methods employing probabilistic treatment of uncertainty. It seems natural to base artificial reasoning systems in these domains on the normative foundations of probability theory. Two usual objections to this approach are (1) probabilistic inference is computationally intractable in the worst case, and (2) probability theory is incomprehensible for humans and, hence, probabilistic systems may be hardly usable. The first objection has been addressed effectively in the last decade by a variety of efficient exact and approximate schemes for probabilistic reasoning, applied in several practical systems. In this paper, I review the state of the art with respect to the second objection.
First I argue that the observed discrepancies between human and probabilistic reasoning and the anticipated difficulties in building user interfaces are not a good reason for rejecting probability theory. On the contrary --- they provide motivation for a normative treatment of uncertainty. I point out that probability theory rests on qualitative foundations that capture essential properties of a domain along with such concepts such as relevance and conflicting evidence. In addition, graphical probabilistic models, as opposed to rule-based systems, integrate numerical and structural properties of a domain and provide a natural representation of causality. Finally, availability of a full quantitative specification of a model allows for manipulating the level of precision for both reasoning and explanation.

Back

Traditional relational or deductive databases can store definite information only. In practical situations, however, much of the information is not precise. Disjunctive deductive databases are logic databases which allow for indefinite or partial information which is formally expressed by means of disjunctions. This paper describes a prototype implementation of such a database system that has been done at the University of Maryland at College Park. We present the main modules of the system and describe optimization techniques used in query evaluation and the storage of indefinite information. We also suggest ways the implemented architecture may be developed into a parallel system.

Back

A combined Gaussian clustering/plug-in estimator of a multivariate probability density, together with a simulation study of its reliability for two- and three-dimensional data, is presented. Interestingly, both the Gaussian clustering algorithm and the plug-in kernel estimator can be considered to be neural networks of some more or less complex forms. Their combination, which is thus a neural network too, is found to be a very promising method of estimating arbitrary continuous densities on $R^d$, at least for small $d$.

Back

In this paper, a neural networks approach to the control of nonlinear dynamic systems is introduced. The advantages and limitations of this approach are discussed. Some fields of control research are indicated as suitable for use in artificial neural networks methods. Also some applications of neural networks as controller and identifier for different technical situations are presented.

Back

Some results of investigations on an adaptive fuzzy expert system are presented. The adaptive fuzzy expert system is constructed as a hybrid in which a fuzzy inference system is combined with a neural network and equipped with a preprocessor of input data, user interface and the decision analyzer. The Liapunov theory is used to examine the non-sensitivity of the optimal value of a generalized weight vector to initial conditions and training data The system is constructed for the needs of an opto-computer system of diagnostic of surface imperfections.

Back

A Cooperative Knowledge-Based System (CKBS) is a collection of autonomous knowledge-based systems called agents which are capable of interacting with each other. A query can be submitted to one agent or a group of agents. An agent when contacted by the user acts as a master agent, he sends requests to other agents which act as his slaves. Clearly, any agent in CKBS can be contacted by a user. So, any agent in CKBS can be a master agent. In this paper, an agent is represented by an information system (either complete or incomplete) and a collection of rules called a knowledge base. Rules we interpret as descriptions of some attribute values in terms of other attribute values. These descriptions are usually not precise and they only provide lower and upper approximations of attribute values. We say that an attribute value is reachable by an agent if either it belongs to the domain of one of the attributes in his information system or it is a decision part of one of the rules in his knowledge base. In the second case all attribute values from the classification part of a rule have to be reachable. Rules in our system are computed at one site of CKBS, sent to other sites of CKBS and stored in their knowledge bases, if needed. So, the set of reachable attribute values at any site of CKBS is constantly changing. Knowledge bases built that way might easily become inconsistent because rules they contain are created independently at different sites of CKBS. The problem of repairing inconsistent rules was investigated in \cite{ras1}. In this paper, we propose a strategy for discovering rules in incomplete information systems as well as we give a method for handling queries in such systems.

Back

The paper concerns the issues of adaptation parameters of evolutionary strategies. Evolutionary strategies together with genetic algorithms and genetic programming create the class of artificial intelligence tools which belong to so called evolutionary algorithms. The evolutionary strategies are characteristic of their self-adjusting properties in response to results achieved during calculations. In this paper we propose further development of the evolutionary strategies by increasing the number of parameters to be changed during execution of the algorithm. The possible areas of applications as well as practical issues of these algorithms will be discussed and a comparison with genetic algorithms will be provided. We consider how to self-adapt the size of population during execution of the algorithm instead of changing only a value of the parameter s as in classical algorithms.

Back

The Generative Policies we will apply to improve performance of the Ant System, are derived from the studies carried in 60th in Germany and USA. First of all we should mention here the leader ( superman ) problem, the \'{e}lite problem and the aggregation problem. We have chosen this problem considering its analogy to the world of real ants. The Ant System we look into, we will confront with the Evolution Computing. First of all we will apply an idea of an evolution of the algorithm itself using the policy connected with one only generation. In conclusion we will propose the optimal values of the parameters used in the Ant System, derived as a result of our experiments.

Back

Grade cluster analysis means clustering based on grade correspondence analysis (GCA) which is a grade version of the classic CA. The ideas and tools of the GCA and the related clustering are shortly reminded in Section 1. In Section 2, GCA is compared with two-dimensional scaling on the basis of the International Morse Code Data, gathered in a psychological experiment conducted in 1957. The ordering of the Morse signals established by the GCA is explained to be a consequence of a few simple rules according to which the subjects participating in the experiment seem to behave. Clusters of adjacent signals are formed and outliers are indicated. Section 3 outlines several examples of GCA and grade clustering performed on various real-life sets of experimental data.

Back

The main task of classification is to predict new data using a set of rules. In this work all rule sets were induced by system LERS (Learning from Examples using Rough Sets). Classification system of LERS has four binary parameters whose values are set before classification. Experiments were done executing this classification system in all 16 possible ways, using nine real-life data sets. The classification method with the smallest error rate was identified.

Back

This publication presents the results of research on methods of processing and graphic presentation of multidimensional analytical data sets and the application of those methods in a system of multicategory objects classification. We have decided to solve those complex problems by using pattern recognition methods. These methods exploit knowledge that is presented in a form of logical associations and recorded in a form of examples to an inference process. From the group of pattern recognition methods we have chosen cluster algorithms, that enable us to define what group of objects the examined object belongs to. Grouping data algorithms enable examination of objects of local area structures, showing also a local structure of a data set. The usefulness of various aggregation algorithms may be estimated while using visual methods, which transform objects of local area from d-dimensions to pictures of 2D- or 3D-dimentions.The results of experiments that were obtained were implemented in program modules of SCANKEE - a computer system that is able to reach a conclusion multistrategically and aid an assistant while choosing the optimal scheme of making a decision.

Back

We discuss problems related to rough relations. We also present modal logic for reasoning about properties of approximations of relations.

Back

The work contains an example of applying the rough sets theory to decide about the necessity of further tests and making final decisions connected with a diagnose, made by a physician, about progressive encephalopathy in a child.
Making the final decisions requires a series of invasive tests and that is why it is essential to carry out processes of appropriate preliminary classification. The process of the preliminary classification of patients with the use of the rough sets theory is presented in the work.

Back

The paper presents a novel view of the Dempster-Shafer belief function as a measure of diversity in relational data bases. It is demonstrated that under the interpretation the Dempster rule of evidence combination corresponds to the join operator of the relational database theory. The interpretation is based on the rough set theory.

Back

The problems of artificial intelligence take significant place in the research for country defense, including the development of military technique. Variety of this research is very wide. Expert system use and development definitely achieved the biggest progress. This was possible due to a relative simplicity in such systems creating as well as low cost of its utilization. However research is much wider and is also concentrated on the sense substitute creating. This mainly concerns the vision (image recognizing and analyzing) and audition (speech recognition). Nowadays computer graphics and speech synthesis are practically well solved.

Back

The paper concerns of a machine visual perception, in particulary, it presents an author's opinion on model-base visual identification systems, using a depth map obtained during a stereo reconstruction proces of a scene on a stereo pair of images, aquired with the aid of pasive stereo cameras.

Back

We present a new constraint logic programming language, Logos, which has an intuitive declarative semantics. Logos computes over an open universe of terms, and returns as answers substitutions constrained by formulas involving disequalities. Logos allows all operators of classical logic including classical negation, existential and universal quantifiers and equality. Moreover it has fully logical input and output facilities. (The final version of Logos is expected to include meta-programming, and have facilities for selection of control which distinguish between sequential and parallel conjunctions of goals.)
Logos is meant as a high level programming language for rapid prototyping in the area of symbolic computation and for construction of rule-based expert systems with logical consistency constraints.
A preliminary version, Logos 0.7 is available via www, both as a Macintosh binary file and a portable source code in Common Lisp.

Back

This paper presents some metodological aspects of designing a model of information structure, information acquisition, and knowledge of abstract intelligent agent. We assume TOGA (Top-down Object-based Goal-oriented Aproach) theory as a fundamental modeling platform for IAW (Intelligent Agent World) and a reference source of terminology. Phenomenological methodology for modeling of some basic components and behaviour of the TOGA Abstract Intelligent Agent (AIA) is suggested. We are focused on information structure, information acquisition, and knowledge of abstract intelligent agent. Phenomenology is epistemic philosophy which, in computer science may be seen as an approach to the generic problems of data acqusition and verification. We would like to present how the phenomenological analysis leads to the confirmation and to some possible extensions of the assumed concepts in TOGA theory. We intend to explain the difference between "natural" and phenomenological point of view while thinking about IAW (Intelligent Agent Worlds). Starting from the phenomenological reduction context of observation we can understand data flow from the d-o-o(domain-of-observation) to AIA as a set of qasi-absolute phenomens. Such modi of IA's observation could help in constructing information and knowledge system embeded in the model of AIA. It seems that, for expressing information structure and knowledge of AIA, the phenomenological point of view could also be helpful for a definition of the adequate domain-independent language. In particular, the suggested fenomenological methodology applied in frame of the TOGA paradigms could be qualitatively new tool in the design of the computer system supporting military strategic decisions.

Back

One of the most important function of natural language is present the reality. The procedures defining the logical structures, what correspond with ones existing in reality, determine the ground to understand the text. The temporal structure is of these structures. On its base, it is possible to elaborate the special "pseudophisical time logic", what axioms answer the peculiarity of perception the time by human being and processs related to it. These kinds of logics comprise very good tool for attainment and arrangement the knowledge in intelligent understing systems of natural language.

Back

This paper describes the methods of program integration of different AI systems within a hybrid system.The first part of this paper describes selected methods of intelligent systems cooperations. The second part presents examples of integration of the expert system and the neural network within the hybrid system PC-Shell.

Back