The performance of a database system significantly depends on the application of suitable indexes. Integrating fuzzy sets in databases requires indexes which can handle that kind of data. One of the few variants concerning an indexing technique for fuzzy data manages the storage of numerical possibility distributions in multidimensional indexes. A subsequent analysis shows that the storage of fuzzy numbers in a B+tree, which is a standard index in most database systems, is the more preferable option. Due to some problematic assumptions in the aforementioned analysis, a reevaluation of that comparison is performed in this paper. In doing so, several modifications of the techniques are proposed and a comparison between B+trees and multidimensional indexes based on these modifications is examined. According to these examinations, a variant of the KDB-tree clearly outperforms the B+tree.

The relatively weak performance of fuzzy database management systems is a major reason for the clear lack of commercial variants. A very important component of a database management system is the query language whose performance is improved amongst others by using indexes. Consequently, the applicability of indexes, designed for the management of fuzzy data, for several operations of a fuzzy query language, in particular a fuzzy equi-join and a fuzzy projection, is examined in this paper. In doing so, the advantages achieved with the integration of a specific index are compared with the benefits resulting from the utilization of a fuzzy equality indicator which is another approach for the efficiency increase of fuzzy query languages. The results will demonstrate that the use of an index can definitely enhance the performance of a fuzzy query language.

Although, in comparison to standard databases, a tremendous benefit is often derived by using fuzzy databases, their distribution is very low. A reason for the relatively poor acceptance of fuzzy relational databases is to be seen in the difficulty to carry out an adequate normalization. The various procedures discussed in the literature possess miscellaneous weaknesses. In this work a normalization method is regarded whose most significant deficit lies in the fuzzy functional dependency (FFD) because thereby comprehensible results are not produced. Therefore, it is registered which alternatives for the determination of the degree of FFD exist. Furthermore, it is examined with which of these methods the just addressed disadvantage can be eliminated. For this purpose, the presented methods are applied to several examples in order to identify their characteristics.

This paper addresses artificial neural network (ANN) evolution by presenting a mutation approach based on a novel self-adaptation strategy. The proposed strategy involves the phenotype information, incorporated in a value, called thenetwork weight(NW), which depends on a total number of hidden layers and an average number of neurons in hidden layers. The inclusion of the phenotype information determines the increment of the mutation step size and the average percentage of successful mutations, which is achieved by means of adaptation to characteristics and the complexity of ANN architectures. The NW operator is combined with the genotype information, included in the dynamic component and represented by the fitness of a particular chromosome. These two components in the mutation approach drive the evolution of chromosomes according to characteristics of an ANN "internal" architecture and a fitness of a particular chromosome simultaneously.

The localization of intracerebral dipole sources in order to detect pathological events is one purpose of magnetoencephalography (MEG). Another aspect is the analysis of brain processes and brain structures. A system consisting of two different types of Artificial Neural Networks is presented. The structure of a feed forward neural network with two layers and a learning rule designed for the task of Blind Signal Separation (BSS) is used to separate temporarily overlapping neuron activities in the brain. Based on the separated signals, the task of the second type of neural network is to determine the position and strength of the different, underlying magnetic dipoles. Several concepts of neural networks, their limits and potentials concerning both tasks of mining medical data are discussed briefly.

Fuzzy controllers are widely used for control purposes. But fuzzy controllers are static, so it is not possible to adapt them or create them automatically. On the other side neural networks are well known for their capability of self-adapting and learning. Therefore it is helpful to represent a given fuzzy-controller by means of a neural network and to have the rules and/or the fuzzy sets adapted by special learning algorithm. Some possibilities in combining fuzzy-controllers with neural networks are the NFECON-model, the model of Lin and Lee or the model we propose: the MFOS (Munsteraner fuzzy optimization system). The MFOS can represent nearly any given fuzzy controller, adapt the rules and fuzzy sets, and transform this optimized net back in a fuzzy-controller. This paper deals with the basic algorithms of the MFOS and more detailed with the fine tuning of the fuzzy rules.

Several ways of combining concepts of fuzzy set theory with connectionist methods are known. We focus on the use of fuzzy numbers in neural networks. Our goal is to create a fully fuzzified Kohonen-layer which receives fuzzy numbers as inputs and computes its output employing fuzzy weights. The main problem is the determination of the winning neuron by the exclusive use of special, "monotonic" fuzzy operations, which guarantee a certain "goodness" of the input/output behaviour. A selection-function is introduced for solving this problem. Furthermore, we formulate a fuzzified version of the standard learning rule, that can be applied on the fuzzified Kohonen neurons.

In classical “crisp” neural networks the output cannot be estimated for arbitrary input data. This situation can be overcome if fuzzy neural nets are trained with fuzzy data. These “continuous” data often better describe certain situations. Because fuzzy neural networks map fuzzy numbers to fuzzy numbers, a criterion for choosing a “good” training set can be formulated. Together with an important fuzzy neural network property, the output for arbitrary crisp input data can be estimated based on the fuzzy training set. © 1998 John Wiley & Sons, Inc.

We present theoretical and empirical investigations to explore the potential of dynamic load balancing for homogeneous systems. We introduce the object-based programming environment MOM. Then we define an own load balancing algorithm called Gating and compare it with other well known algorithms, theoretically, in a simulation and in the MOM system. The obtaining results are discussed.

This paper presents some recent results in interpreter optimization. The techniques of shallow binding and repetitive interpretation of tail recursive functions are adapted to Lisp with static scoping as the binding method for-all identifiers. Then a new technique of interpreting" covered tail recursive" functions is proposed. The purpose of the paper is to show that the extra expense for static scoping can be kept small by combining these techniques.