We discuss the issue of scheduling agents from two aspects: (1) how the concept and technology of intelligent agents can help to model and solve problems of scheduling and, on the other way around, (2) what are the prospects of applying scheduling methods in the coordination of multi-agent systems. Focusing on production scheduling, we analyze the gap between theory and practice, argue for a novel modeling approach, and show that many of the new, mostly practical requirements can be met by autonomous, reactive and distributed agents that are embedded in their execution environment.

Agents endowed with such properties are appropriate not only for capturing relatively fixed resources -- which is quite usual -- but also for modeling items undergoing production. It turns out that this latter, in fact, can help exploring the possibilities of the second aspect (2) of our target.

Intuitively, mobile agents provide the means for realizing items during production, i.e., agents "on the go". However, the application of mobile agents is not without risks: we attempt to determine the motives that call for and the conditions that make possible their use in scheduling.

We propose a formal semantics for Machine Learning, particularly for Learning Agents. We start from the classical definitions in the domain. The basic point is to state Learning as the building of a visibility relationship between possible worlds. We try to use in this framework the basic idea of emergence, which is to build an upper level which has a causal influence on the lower one.

Agent based approach provides a new paradigm for analyzing, designing and implementing complex systems. The main feature of agent oriented modeling is to view systems as a set of interacting agents within an environment. This provides a natural and intuitive approach to model a broad variety of applications such as Artificial Life, Collective Robotics, Distributed Problem Solving and Simulation. Although many Multi-Agent Systems (MAS for short) have been designed, there is a crucial lack concerning design and development methodologies. A process of specification is fundamental to handle the complexity related to building such systems and specifying the desirable behavior of MAS before their implementation phase. The specification process must fulfill two roles. The first is to provide the underlying rationale for the system under development. The second is to guide subsequent design, implementation and verification phases. A variety of specification formalisms are available in the multi-agent field. Such formalisms put the emphasis on the first role and do not provide a basis to fulfill the second as they are often abstract and unrelated to concrete computational models. We believe that one way to bridge the gap between the abstract and concrete level is to build the specification of systems using a prototyping process. This process provides a support for incremental specification leading to an executable model of the system being built. Indeed, in many areas of software and knowledge engineering, the development process putting emphasis on prototyping and simulation of complex systems before their effective implementation is proven to be a valuable approach. The purpose of this talk is to present several fundamental issues relating to methods and formalisms for the specification and development of multi-agent systems.A specification method is essential to manage MAS complexity by decomposition and abstraction. We also present an organizational model based on organization, interaction and role to specify MAS. It allows to go from requirements to a detailed design and helps to decompose a MAS in terms of roles and organizations.

The paper is concerned with distributed and parallel discrete simulation which offers the possibility of executing time-consuming calculations on computer networks or parallel computers. Distributed event-driven simulation is proposed as an alternative to traditional sequential simulation. The paper discusses some important issues associated with the implementation of distributed simulation. A particular attention is focused on the software environment which provide framework for simulation experiments performed on parallel computers. The paper presents the software system CSA&S-PV (Complex Systems Analysis & Simulation - Parallel Version) designed as a software tool which supports the developers of the control mechanisms for large scale systems. The program allows to perform multiple simulation experiments using parallel asynchronous computing. The CSA&S-PV system was used to investigate several real-life problems. The application is presented for flow transformation in the river basin and flood control in multiple reservoir system.

The process of cooperative problem solving can be divided into four stages. First, finding potential team members, then forming a team followed by constructing a plan for that team. Finally, the plan is executed by the team. Traditionally, very simple protocols like the Contract Net protocol are used for performing the first two stages of the process. In an open environment however, there can be discussion among the agents in order to form a team that can achieve the collective intention of solving the problem. For these cases fixed protocols like contract net do not suffice. In this paper we present an alternative solution, using structured dialogues, with an emphasis on persuasion, that can be shown to lead to the required team formation. The dialogues are described formally using modal logics and speech acts.

The message system considered in the paper consists of a finite set of places, each of them capable to store a fixed number of messages. Places can be joined by links and then they are neighbours. Messages are created and moved from places to neighbour places until reaching their destinations and then disappear. The system is supposed to act in a non-terminating way: newly created messages are flowing through the network and vanishes after reaching their destinations, making room for creation new messages. Messages are moving through the network independently of each other; however, because of limited capacity of places, some messages can stand in the way of reaching destination by others; in effect of such obstacles some messages may not reach their target at all. The aim of this paper is to define a rule of message moving which ensures responsiveness of the system, i.e. a rule that guarantees any message appearing in the system to reach eventually its destination. In particular, such a rule should prevent messages from endlessly circulating in the network, or from being constantly hampered by others. Such a rule, based on so-called "potential function" of messages, is found and its adequacy is proved.

Evolutionary algorithms are kind of stochastic iteration schemes uses mostly to solve difficult optimization problems. In this lecture we discuss the underlying Markovian models, the concept of convergence to the global optimum, sufficient criteria for convergence and measures of convergence rate. A special attention is devoted to the finite-state models.

The paper proposes using genetic algorithms - based learning classifier system (CS) to solve multiprocessor scheduling problem. For this purpose, agents - local decision making units are associated with corresponding tasks of a parallel program, and the program is interpreted as a multi-agent system. After initial mapping tasks into processors of a parallel system, the agents perform migration to find an allocation providing the minimal execution time of the program. Decisions concerning agents' actions are produced by the CS, upon a presentation by an agent information about its current situation. The CS is a rule - based system which is able to learn rules of scheduling during its operation. Actions of agents can be executed by them sequentially, in a group or in parallel. Results of experimental study of the scheduler are presented.

Spatial reasoning comprises reasoning under uncertainty about spatial concepts. Usually, schemes for spatial reasoning make use of ontological taxonomy of concepts and a mereological calculi based mostly on the notion of a connection. We will present an alternative approach to spatial reasoning based on rough mereology, a paradigm for reasoning under uncertainty extending mereology and especially fit for special data encoded in spatial information systems. We will demonstrate how basic constructs for spatial reasoning may be derived in the rough mereological environment and how they may be synthesized by distributed or many-agent systems.

Relatively large number of agents in mass multi-agent systems (mMAS) stems necessity of the new approach to analysis, design and utilisation of such systems. Their double nature: micro (virtual) -- with respect to internal phenomena among agents, and macro (real) -- which arises at the interface to the real world, must be thoroughly considered. In the paper the V-R decomposition is proposed as the idea that structures the problem. As an illustration, some results of simulation studies on the evolutionary predicting system -- an example of mMAS -- are presented and discussed.

We discuss the mathematical model of both architecture and learning process of artificial neural networks. Dynamical systems theory is used to describe the learning process of networks consisting of linear, weakly nonlinear and nonlinear neurons. Conjugacy between a gradient dynamical systen wth a constant time step and a cascade generated by its Euler method is applied as well.

In this paper, we present quantitative estimation of heterogeneous distributed computing systems. It was shown that known methods for general estimation of heterogeneous distributed computing systems as usual allow to estimate average deviation for nodes characteristics of distributed computing systems from maximal or minimal estimation of computing node. We exploit analytical expression which allows to make the quality-quantity general estimation of the heterogeneous degree for heterogeneous distributed computing system. We consider six different methods for measuring of heterogeneous degree of heterogeneous distributed system. We propose model of heterogeneous computing system and apply it to calculation of heterogeneous degree considering system’s characteristics. Experimental results show that our model can provide parameters of computing system with more accuracy and consider not only quantitative characteristics of computing nodes but their quality structure as well.

In this paper, we consider the problems of time optimization of big data arrays processing in distributed multiprocessors systems. This optimization is based on analysis of the time spending for data interchange between computer nodes. We include a comparison of two methods of data distributing. This comparison shows that the method of preliminary data distributing provides considerable decrease of execution time. Moreover, at the stage of task mapping the method of preliminary data distributing considers of executing speeds of each processors that leads to efficient processors loading. We also propose an analytical model that allows to estimate the volume of data sending between nodes and optimization of whole execution time.

Background of the research
Computer networks offer new application scenarios that cannot be realized on a single workstation. However, for large computer networks like the cyberspace (the open world created by the global information infrastructure and facilitated by the Internet and the Web) the classical programming paradigm is not sufficient. It is hard to imagine and even harder to realize the control, synchronization and cooperation of hundreds of processes running on remote hosts.
The idea of mobile software agents inhabiting the cyberspace seems to be a good solution here. A mobile agent is an executing program that can migrate from host to host across a heterogeneous network under its own control and interact with other agents. However, it is clear that a single agent cannot perform efficiently its tasks in a large open world without cooperation with other agents.
For large open worlds simple cooperation mechanisms based on bilateral communication are not sufficient. Sophisticated interaction mechanisms and services are needed. During the interactions agents communicate, negotiate, and form organization structures. To realize the concept of mobile agents together with agents interactions and service infrastructure, a special middleware called "mobile agent platform" (MAP, for short) is needed.
There are several platforms available over the Internet, for example IBM Agelets, Concordia, Grasshopper, Mole, and Voyager to mention only some of them. One of them is Pegaz developed in our Institute. These platforms are for creating infrastructures on computer networks, so that details of network functioning are hidden from the users as well as from the agents, see Fig. 1. This makes programming more easy. A programmer need not manually construct agent communication, nor agent transportation. This may be viewed as high level abstraction from network communication protocols, operation systems (due to Java), and data structures.
Most of the existing mobile agent platforms create similar infrastructures. However, there is a problem of interoperability between different platforms if we want agents to migrate from one platform to another. Based on OMG CORBA, MASIF standard [8] offers interoperability limited to implementation independent aspects. Agents that migrate from platform A to platform B must know how to access the internal services and resources offered by the platform B. Such object-internal aspects are usually not handled by CORBA.
Since most of today's mobile agent platforms are implemented in Java, it is postulated in [7] to create a new OMG standard to define implementation-specific (i.e. Java-specific) convention that allows a mobile agent to migrate to any standard-compilant platform. This concept (a specification of a generic mobile agent platform) has been developed in GMD FOCUS and IKV++ [3], and realized in the Grasshopper platform [4].
However, the interoperability limited only to the core functionality offered by the Grasshopper architecture is not sufficient for a mobile agent to be able to access internal services and resources (of a remote place) being beyond the scope of the core functionality. To do so the agent must "know" how and for what purpose the services and resources can be used, or at least to be able to learn about that.
In order to assure this, a common language based on the common ontology is needed. The language is also necessary for defining agent negotiation protocols, joint plans, and intentions. According to the standard definition, an ontology is an explicit specification of a conceptualization, see [5]. A conceptualization is an abstract, simplified view of the world. Hence, in order to define an ontology and a language, a formal representation of the world is needed. In our case, the world is the environment created by mobile agent platforms. Let the environment be called cyberspace. Since most of the existing MAPs create similar infrastructures, it seems that a formal specification of the cyberspace is possible. In our opinion the common ontology based on a formal specification of the cyberspace is a way to achieve a high interoperability level.
There is another effort taken by FIPA [2]. This approach to ontology is based on a different point of view. It is supposed that each domain (for example, e-commerce) has its own specification expressed in a formal language. This specification is identified with the ontology of the domain. So that the ontology provides a vocabulary for representing and communicating knowledge about the domain. If two agents wish to converse, they must share a common ontology for the domain of discourse. Quite recently DARPA has proposed another approach called DARPA Agent Markup Language (DAML) [1]. DALM is to be a language (built on XML) that could serve as means for software agents to dynamically identify, communicate and understand each other. " ... Going far beyond XML, the goal of this program is to develop a language aimed at representing semantic relations in machine readable ways compatible with current and future Internet technologies. ..."
Our approach is a bit different. We construct a representation of cyberspace (understood as the environment created by MAPs) as the basis for constructing application domain ontology. This very representation serves also as the basis for constructing the key aspects of agent architecture, i.e., common ontology, agent knowledge and a simple agent communication language. Our approach to ontology follows the idea of Wittgenstein that the meaning of language is in its use. In the approach of Gruber, et al. [5] and Gaurino [6], the interpretation (meaning) of concepts is constrained by logical axioms. Equipping agents with goals and decision making mechanisms, we can construct multi-agent systems. We also develop a formal theory (based on modal logic) of such multi-agent systems.
The research is based on our experiences gained during realization of Peagz - our MAP, and modeling agent virtual organizations as sophisticated agent interaction mechanisms (see www.ipipan.waw.pl/mas/).

Modeling many real life problems requires solving complex systems of difference equations. One of the most useful methods which is applied to solving such problems is a finite element analysis (FEA). In this method, the system of difference equations is finally reduced to a linear system, created for finite number of elements or nodes. Finite element mesh is a part of finite element model which is created for every domain concerned by a problem.
Generating the finite element mesh is one of the most important tasks in building the finite element model. Although there are a lot of algorithms developed, very often meshes are poorly shaped, so the smoothing process is necessary. Smoothing is a technique designed for element shape optimization. There are many methods of smoothing, but it is hard to describe a method which would be satisfactory enough for every kind of meshes.
Many applications of FEA have too large computing or/and memory cost for a sequential implementation. Parallel compuations allow overcoming these bottleneck by dividing a finite element mesh into a number of sub-domains (or sub-meshes) which can be processed concurrently over different processors. As a result, the mesh partitoning is a vital part of implementing the finite element model in parallel. The task of partitioning unstructured meshes for parallel finite element computations is not straightforward, since not only the load per sub-domains has to be kept the same. Also, interprocessor communications have to be minimized. These communications are results of interaction (or coupling) between boundary nodes of sub-meshes.
Genetic algorithms (GAs), based on mechanisms of natural selection and inheritance, are known to be one of the best optimization methods. Thanks to their stability, they can be applied to searching into even very complex search spaces.
This work deals with applications of GAs in finite element computations. Two areas are considered: mesh smoothing, as well as parallel mesh partitioning and generating.
The goal of finite element shape optimization is to find such coordinates of nodes that provide the most regular shape of the element containing these nodes. The regularity of element shape is necessary for the minimization of discretization errors. For triangular meshes, the ideal shape of element is the equilateral triangle. The optimization algorithm searches for a new position of a node, and only one node is moved in an iteration. To solve the optimization problem, the compact GA (shortly cGA) is used. It differs from the classical GA in the population representation and population processing. The important feature of cGA is that it needs significantly less memory than the classical GA.
In the paper, an object-oriented software implementation of the algorithm developed is presented, as well as some results of experiments with mesh smoothing carried out on the SUN Ultra 10/300 workstation. These results show that the cGA algorithm can be used even for optimization of very poor-shaped meshes; however, in this case, it is very time-consuming. The reason is that the algorithm runs many iterations for each optimized node. The goal of future works is to include the possibility of displacement of boundary nodes. Another idea is to expand the proposed algorithm in such a way that all nodes of an element will be relocated simultaneously. In this case, a single chromosome will code positions of several nodes.
When solving the second problem, namely, the finite element mesh partitioning and generating, it is supposed that a coarse background mesh with data describing the density in the final mesh are given. There are two possible approaches to the problem of mesh partitioning and refining. One is to implement two consecutive steps of refining and partitioning on a single processor, and then send the obtained results to remaining processors. In the opposite case, refining is made on target processors concurrently. The second approach, which is investigated in the paper, allows avoiding memory and time bottlenecks, which may take place in the first approach. In the method presented in the paper, a genetic algorithm with Gray coding and tournament selection is used to find coordinates of the dividing vector in the basic two- subdomain partitioning procedure. This procedure is recursively repeated for created sub- domains, and the number of resulting sub-domains depends on the number of available processors.

Mobile robots are well suited for reconnaissance and surveillance. Depending upon needs, such robots can detect technological hazards (fire, chemical or nuclear contamination, unac- ceptable temperature or humidity, etc.) as well as the presence of intruder. The aim of the pre- sent paper is to apply the multi-agent paradigm to such tasks.
For practical reasons, we have chosen the mixed model in which both hierarchy and auton- omy are present. We treat a certain number of mobile robots as a team of agents. One of them is distinguished as a leader. The team members communicate with each other via wireless links. The leader communicates with the human operator. The latter initiates the task for the team of robots and if necessary corrects their decisions.
The behavior of each agent is modeled at 2 levels. A scenario describes the behavior (tactics) of the whole team under certain circumstances. A role is assigned in such a scenario for a particular agent. Such a role defines actions undertaken by the robot as well as gives the mode of its co-operation with other team members. At present we deal with a fixed list of scenarios and roles. In the future an ability to learn them will be introduced.
In order to test our concept, we developed a simulator in Java. This application allows the user to describe the map of environment, to form a team of agents, to select a task to be solved by them and to observe the behavior of agents. A problem of finding a given object in the building has been chosen as a test. The results will be presented at the conference.

We present the new evolutionary strategy which gives good computation complexity in solving global optimisation problems. A basic ideas of this strategy are hierarchical decomposition and utilization of different length genotypes. This algorithm exhibits its efficiency especially in case of many solutions with distant basins of attraction.