摘要:

The results to date are presented from an ongoing investigation, in which the aim is to combine the strengths of different artificial intelligence methods into a single problem solving system. The premise underlying this research is that a system which embodies several cooperating problem solving methods will be capable of achieving better performance than if only a single method were employed. The work has so far concentrated on the combination of case-based reasoning and artificial neural networks. The relative merits of artificial neural networks and case-based reasoning problem solving paradigms, and their combination are discussed. The integration of these two AI problem solving methods in a hybrid systems architecture, such that the neural network provides support for learning from past experience in the case-based reasoning cycle, is then presented. The approach has been applied to the task of forecasting the variation of physical parameters of the ocean. Results obtained so far from tests carried out in the dynamic oceanic environment are presented. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58249

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

As the size of typical industrial strength information systems continues to rise, particularly in the arena of Internet based management information systems and multimedia servers, the issue of managing data distribution over clusters or ‘farms’ to overcome performance and scalability issues is becoming of paramount importance. Further, where access is global, this can cause points of geographically localized load contention to ‘follow the sun’ during the day. Traditional site mirroring is not overly effective in addressing this contention and so a more dynamic approach is being investigated to tackle load balancing. The general objective is to manage a self-adapting, distributed database so as to reliably and consistently provide near optimal performance as perceived by client applications. Such a management system must be ultimately capable of operating over a range of time varying usage profiles and fault scenarios, incorporate considerations for communications network delays, multiple updates and maintenance operations. It must also be shown to be capable of being scaled in a practical fashion to ever larger sized networks and databases. Two key components of such an automated system are an optimiser capable of efficiently finding new configuration options, and a suitable model of the system capable of accurately reflecting the performance (or any other required quality of service metric) of the real world system. As conditions change in the real world system, these are fed into the model. The optimiser is then run to find new configurations which are tested in the model prior to implementation in the real world. The model therefore forms an evaluation function which the optimiser utilises to direct its search. Whilst it has already been shown that Genetic Algorithms can provide good solutions to this problem, there are a number of issues associated with this approach. In particular, for industrial strength applications, it must be shown that the GA employed can provide reliable and consistent performance. This paper investigates evolutionary computation techniques, comparing results from genetic algorithms, simulated annealing and hillclimbing. Major differential algorithm performance is found across different fitness criteria. Preliminary conclusions are that a genetic algorithm approach seems superior to hillclimbing or simulated annealing when more realistic (from a quality of service viewpoint) objective functions are used. Further, the genetic algorithm approach displays regions of adequate robustness to parameter variation, which is also critical from a maintained quality of service viewpoint. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58264

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

The paper refers to earlier results obtained by the authors and constitutes their essential complement and extension by introducing to a diagnostic model the assumption that the decision concerning the diagnosis is based on observations of symptoms carried out repeatedly, by stages, which may have effect in a change of these symptoms in increasing time. The model concerns the observations of symptoms at an individual patient at a time interval. The changes of the symptoms give some additional information, sometimes very important in the diagnostic process when the clinical picture of a patient in a certain interval of time differs from that one which has been received from the beginning of the disease. It may occur that the change in the intensity of a symptom decides an acceptance of another diagnosis after some time when the patient does not feel better. The aim is to fix an optimal diagnosis on the basis of clinical symptoms typical of several morbid units with respect to the changes of these symptoms in time. In order to solve such a posed problem the authors apply the method of fuzzy relation equations which are modelled by means of logical rules of inference. Moreover, in the final decision concerning the choice of a proper diagnosis, a normed Euclidean distance is introduced as a measure between a real decision and an “ideal” decision. A simple example presents the practical action of the method to show its relevance to a possible user. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58266

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

Conceptual structures like conceptual graphs and concept lattices are formal representations of knowledge, e.g. structure or information content. In this paper, it is argued that, based on Peirce’s semeiotic theory of signs and linguistic knowledge about syntactic clusters, conceptual structures can be derived in natural language automatically. This is incorporated in a novel model of language called Natural Language Concept Analysis (NLCA). ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58267

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

In the following we discuss two models of ecological- or biological-like evolution. First, a simple dynamical model studied through Monte Carlo simulations and describing the gradual evolution of species, is mainly emphasized: there are two independent competing mechanisms for evolution or growth, i.e. two for removing a species from the system, and one for creating a new species. Regardless of the initial concentration of species,the system always reaches the same final state, characterized by the same concentration and the same average adaptation. The system is finally not homogeneous and contains species with different values of the adaptation. The evolution does not show avalanches, nor punctuated equilibria, but leads to an organized state. The behavior seems to have been found in evolving biological systems. The next simple dynamical model is based on the Bak-Sneppen model, but with a variable number of species as done by Vandewalle and Ausloos in 1996, allowing for branching, death of ancestor, and “interaction” between the new species at finite or infinite range. The system evolution shows avalanches, punctuated equilibria, and screening of evolution within an organized state. The geometrical, biological and physical critical exponents are numerically determined and lead to simple relations among them. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58268

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

The principles of the emergence of new qualities are analyzed on the basis of the evolutionary analogies of biological systems and models of interacting systems. It is pointed out thatcompetitiondoes not create new qualities; it only elaborates on and disseminates a certain trait of the system. New qualities are generated by symbiotic interactions:mutualism, cooperationandpredator-prey-likeones. Special attention is called to the evolution of the predator-prey-like system, as it evolves into anorganizedand evenanticipatorysystem: when this system becomes a simple cyberneticregulator, when the predator specializes to process ∼1 bit of information; when the system can regulate many parameters and the predator becomes a complexprocessorof information that controls the activity of the Prey, that is the transformations of matter/energy, when additional memory structures that contain fixed sets of programs emerge (programmed control); and when, finally, a special structure that canmodelthe internal and external world andstore informationdevelops, and the whole system becomes ananticipatory system. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58265

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

A recently introduced model of macroevolution is studied on two different levels of systems analysis. Firstly, the systems dynamics and properties, above all the growth of complexity of the evolutionary units during the long-term evolution, are discussed, and, secondly, the complexity of the model itself, i.e. the richness of its various features, is studied with regard to a control parameter representing a background noise within the systems dynamics. The same is done with a randomized version of the model. The model is based on a normalized one-dimensional coupled map lattice with locally interacting sites representing different species. The evolution of the sites’ values representing the fitness of the species is governed by a usual diffusion rule and an additional memory- or random-based feedback loop. The introduction of a realistic background noise limiting the range of the feedback operation yields a pattern signature in fitness space with a distribution of temporal boost/mutation distances similar to a punctuated equilibrium behavior. Furthermore, the behavior of the mean lifetimes of “high” fitness values is correlated with the resolution-like parameter &Vegr; via a power law, a phenomenon called “fractal evolution.” Based on simple functional properties of the power law, an additional feedback loop is introduced to use the intrinsic fluctuations of the whole fitness landscape as a driving force to change adaptively the systems resolution. On long-term scales, the dynamical system properties exhibit a clear tendency towards progressive evolution potentials for each species. For both model versions, the memory-based and the random-based one, we achieve some basic mechanisms of evolutionary dynamics like coevolution, punctuated equilibrium with regard to internal or external changes during evolution, coordinated stasis for groups of species, and self-organized growth of complexity for all evolutionary units of the array leading to a kind of “Red-Queen-effect.” Additionally, for the memory based model a parameter was found indicating a limited range of noise allowing for the most complex behavior of the model, whereas the entropy of the system provides only a monotonous measure with respect to the varying noise level. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58250

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

When a dynamical system is studied, one of the most interesting goals is to obtain a set of differential equations which provides a model of its behavior. Indeed, it has been pointed out that the knowledge of a global model which captures all the underlying dynamics provides a significant step in the understanding of physical processes. In the case of experiments, the aim is therefore to obtain a set of ordinary differential equations which generates a behavior equivalent to the experimental one starting from a scalar time series given for an observable. The method, the so-called global vector field reconstruction, is described as well as exemplified on experimental cases arising from chemistry, namely a copper electrodissolution and a Belousov-Zhabotinskii reaction. Numerical examples for taking into account a control parameter or for modelling a driven system are also discussed. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58270

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

A review of the approach to the study of turbulence (spatiotemporal chaos) in a finite spatial region, based on decomposition of the underlying partial differential equations over an infinite set of modes and subsequent truncation of the infinite chain of the corresponding ordinary differential equations (ODEs), is given. The review starts with a general survey of this approach. Then, to illustrate it in terms of real systems, two recently studied models are considered in some detail. The first model is a system of six ODEs for amplitudes of eigenmodes of the curl operator, which is a simplest nontrivial truncation of the three-dimensional (3D) Euler equations; the wave vectors of the six modes are closed into a simplex (tetrahedron) in the 3D space, which makes this truncation basis very natural. The resultant dynamical system (DS) has three integrals of motion, that can be interpreted as the energy, helicity, and squared angular momentum of the flow. Depending on values of these conserved quantities, numerical simulations reveal both regions of purely chaotic (ergodic) motion, and mixed regions where chaotic and regular motions coexist. The second model is produced by truncation of the two-dimensional (2D) complex Ginzburg-Landau (CGL) equation in a rectangular region, the spatiotemporal chaos being triggered by the modulational instability (MI) of a spatially uniform solution. Close to the MI threshold, the instability is dominated by few spatial modes, which makes it possible to perform the truncation in a consistent way. This leads to a fifth-order DS, that admits two different invariant reductions to third-order DSs. One reduction corresponds to a particular case of the 1D CGL equation, and it this case the system can be further transformed into the famous Lorenz system or its modified form. The other reduction produces DS that has coexisting exploding (singular) solutions and dynamical chaos accounted for by a compact strange attractor. Both types of the solutions are meaningful, the singular ones corresponding to the wave collapse in the 2D CGL equation. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58271

出版商:AIP    

年代:1999

数据来源: AIP

摘要:

Processing of strings by some automata, when viewed on space-time (ST) diagrams, reveals characteristic soliton-like coherent periodic objects. They are inherently associated with iterations of automata mappings thus we call them the iterons. In the paper we present two classes of one-dimensional iterons: particles and filtrons. The particles are typical for parallel (cellular) processing, while filtrons, introduced in (32) are specific for serial processing of strings. In general, the images of iterated automata mappings exhibit not only coherent entities but also the fractals, and quasi-periodic and chaotic dynamics. We show typical images of such computations: fractals, multiplication by a number, and addition of binary numbers defined by a Turing machine. Then, the particles are presented as iterons generated by cellular automata in three computations: B/U code conversion (13, 29), majority classification (9), and in discrete version of the FPU (Fermi-Pasta-Ulam) dynamics (7, 23). We disclose particles by a technique of combinational recoding of ST diagrams (as opposed to sequential recoding). Subsequently, we recall the recursive filters based on FCA (filter cellular automata) window operators, and considered by Park (26), Ablowitz (1), Fokas (11), Fuchssteiner (12), Bruschi (5) and Jiang (20). We present the automata equivalents to these filters (33). Some of them belong to the class of filter automata introduced in (30). We also define and illustrate some properties of filtrons. Contrary to particles, the filtrons interact nonlocally in the sense that distant symbols may influence one another. Thus their interactions are very unusual. Some examples have been given in (32). Here we show new examples of filtron phenomena: multifiltron solitonic collisions, attracting and repelling filtrons, trapped bouncing filtrons (which behave like a resonance cavity) and quasi filtrons. ©1999 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.58241

出版商:AIP    

年代:1999

数据来源: AIP