Usually, probabilistic automata and probabilistic grammars have crisp symbols as inputs, which can be viewed as the formal models of computing with values. In this paper, we first introduce probabilistic automata and probabilistic grammars for computing with (some special) words in a probabilistic framework, where the words are interpreted as probabilistic distributions or possibility distributions over a set of crisp symbols. By probabilistic conditioning, we then establish a retraction principle from computing with words to computing with values for handling crisp inputs and a generalized extension principle from computing with words to computing with all words for handling arbitrary inputs. These principles show that computing with values and computing with all words can be respectively implemented by computing with some special words. To compare the transition probabilities of two near inputs, we also examine some analytical properties of the transition probability functions of generalized extensions. Moreover, the retractions and the generalized extensions are shown to be equivalence-preserving. Finally, we clarify some relationships among the retractions, the generalized extensions, and the extensions studied recently by Qiu and Wang.

Through the Internet and the World-Wide Web, a vast number of information sources has become available, which offer information on various subjects by different providers, often in heterogeneous formats. This calls for tools and methods for building an advanced information-processing infrastructure. One issue in this area is the selection of suitable information sources in query answering. In this paper, we present a knowledge-based approach to this problem, in the setting where one among a set of information sources (prototypically, data repositories) should be selected for evaluating a user query. We use extended logic programs (ELPs) to represent rich descriptions of the information sources, an underlying domain theory, and user queries in a formal query language (here, XML-QL, but other languages can be handled as well). Moreover, we use ELPs for declarative query analysis and generation of a query description. Central to our approach are declarative source-selection programs, for which we define syntax and semantics. Due to the structured nature of the considered data items, the semantics of such programs must carefully respect implicit context information in source-selection rules, and furthermore combine it with possible user preferences. A prototype implementation of our approach has been realized exploiting the DLV KR system and its plp front-end for prioritized ELPs. We describe a representative example involving specific movie databases, and report about experimental results.

We investigate knowledge exchange among commercial organi sations, the rationale behind it and its effects on the marke t. Knowledge exchange is known to be beneficial for industry, bu t in order to explain it, authors have used high level concept s like network effects, reputation and trust. We attempt to formal ise a plausible and elegant explanation of how and why compan ies adopt information exchange and why it benefits the market as a whole when this happens. This explanation is based on a multi - agent model that simulates a market of software providers. E ven though the model does not include any high-level concept s, information exchange naturally emerges during simulation s as a successful profitable behaviour. The conclusions reac hed by this agent-based analysis are twofold: (1) A straightforward se t of assumptions is enough to give rise to exchange in a softwa re market. This work was carried out when M. Chli and P . The growth of the Internet as a medium of knowledge exchange has stimulated a lot of scientific interest origina ting from various disciplines. The willingness of individua ls, organisations as well as commercial firms to share information via the Internet has been remarkable. In some sectors like scientific research, the communication of newly acquir ed knowledge and expertise in a field is considered vital for the ir advancement. On the other hand, in other sectors, the benefit s of such exchanges may not be obvious. For instance, it might even be considered damaging for pharmaceutical companies t o make public any innovations generated by their Research and Development (R&D) process. In spite of this view, exchange o f intellectual property in some industries occurs quite freq uently and in various different ways. These include the forming of strategic partnerships, the participation in open source s oftware projects and the publication of scientific papers by researc h labs that are part of commercial companies. W e study the knowledge exchange that occurs in the software industry. In particular, we focus on analysing the rationale behind this exchange as well as its effect on the industry. The complexity of software requirements is a char - acteristic that distinguishes the software market from oth ers.

The adaptation rule for Vector Quantization algorithms, and consequently the convergence of the generated sequence, depends on the existence and properties of a function called the energy function, defined on a topological manifold. Our aim is to investigate the conditions of existence of such a function for a class of algorithms examplified by the initial ''K-means'' and Kohonen algorithms. The results presented here supplement previous studies and show that the energy function is not always a potential but at least the uniform limit of a series of potential functions which we call a pseudo-potential. Our work also shows that a large number of existing vector quantization algorithms developped by the Artificial Neural Networks community fall into this category. The framework we define opens the way to study the convergence of all the corresponding adaptation rules at once, and a theorem gives promising insights in that direction. We also demonstrate that the ''K-means'' energy function is a pseudo-potential but not a potential in general. Consequently, the energy function associated to the ''Neural-Gas'' is not a potential in general.

Department of Mathematics, University of New Mexico, Gallu p, NM 87301, U.S.A. Abstract: This paper presents two new promising combination rules for the fusion of uncertain and potentially highl y conflicting sources of evidences in the theory of belief func - tions established first in Dempster-Shafer Theory (DST) and then recently extended in Dezert-Smarandache Theory (DSmT). Our work is to provide here new issues to palliate the well-known limitations of Dempster's rule and to work beyond its limits of applicability. Since the famous Zadeh' s criticism of Dempster's rule in 1979, many researchers have proposed new interesting alternative rules of combination to palliate the weakness of Dempster's rule in order to provide acceptable results specially in highly conflicting situati ons. Bot h rules allow to deal with highly conflicting sources for stati c and dynamic fusion applications. W e present some interesting properties for ACR and PCR rules and discuss some simulation results obtained with both rules for Zadeh's pro b-lem and for a target identification problem.

In this short paper we present a linear constraint solver for the UniCalc system, an environment for reliable solution of mathematical modeling problems.

Although the Internet AS-level topology has been extensively studied over the past few years, little is known about the details of the AS taxonomy. An AS "node" can represent a wide variety of organizations, e.g., large ISP, or small private business, university, with vastly different network characteristics, external connectivity patterns, network growth tendencies, and other properties that we can hardly neglect while working on veracious Internet representations in simulation environments. In this paper, we introduce a radically new approach based on machine learning techniques to map all the ASes in the Internet into a natural AS taxonomy. We successfully classify 95.3% of ASes with expected accuracy of 78.1%. We release to the community the AS-level topology dataset augmented with: 1) the AS taxonomy information and 2) the set of AS attributes we used to classify ASes. We believe that this dataset will serve as an invaluable addition to further understanding of the structure and evolution of the Internet.

A model of an organism as an au tonomous intelligent system has been proposed. This model was used to analyz e learning of an organism in various environmental conditions. Processes of learning were divided into two types: strong and weak processes taking place in the absence an d the presence of aprioristic information about an object respectively. Weak lear ning is synonymous to adaptation when aprioristic programs already available in a system (an organism) are started. It was shown that strong learning is impossible fo r both an organism and any autonomous intelligent system. It was shown also that the knowledge base of an organism cannot be updated. Therefore, all behavior programs of an organism are congenital. A model of a conditioned reflex as a series of consecutive measurements of environmental parameters has been advanced. Repeated measurements are necessary in this case to reduce the error during decision making.

Physical processes are computations only when we use them to externalize thought. Computation is the performance of one or more fixed processes within a contingent environment. We reformula te the Church-Turing thesis so that it applies to programs rather than to c omputability. When suitably formulated agent-based computing in an open, multi-scalar environment represents the current consensus view of how we interact with the world. But we don't know how to formulate multi-scalar environments. Keywords: agents, agent-based, agent-based computation, Church-Turing thesis, Church's thesis, computing, computation, envir onment ideas, interaction, interactive computation, models, multi-scalar envir onment, thought, thought tools, unconventional computation.

We address the problem of reinforcement learning in which observations may exhibit an arbitrary form of stochastic dependence on past observations and actions. The task for an agent is to attain the best possible asymptotic reward where the true generating environment is unknown but belongs to a known countable family of environments. We find some sufficient conditions on the class of environments under which an agent exists which attains the best asymptotic reward for any environment in the class. We analyze how tight these conditions are and how they relate to different probabilistic assumptions known in reinforcement learning and related fields, such as Markov Decision Processes and mixing conditions.