Robert Nowicki

On design of flexible neuro-fuzzy systems for nonlinear modelling

Abstract In this work, we consider the flexible neuro-fuzzy systems of the Mamdani-type. When designing such systems to solve approximation problem, we should choose triangular norms used in inference and aggregation operators. This can be done by trial and error. In this work, we propose an algorithm that allows in an automatic way to choose the types of triangular norms in the learning process. The task of this algorithm is also an automatic selection of parameters of all functions describing the system. The algorithm uses an evolutionary strategy for its action and has been tested using well-known approximation benchmarks.

Rough Neuro-Fuzzy Structures for Classification With Missing Data

This paper presents a new approach to fuzzy classification in the case of missing data. The rough fuzzy sets are incorporated into Mamdani-type neuro-fuzzy structures, and the rough neuro-fuzzy classifier is derived. Theorems that allow the determination of the structure of a rough neuro-fuzzy classifier are given. Several experiments illustrating the performance of the rough neuro-fuzzy classifier working in the case of missing features are described.

On Combining Neuro-Fuzzy Architectures with the Rough Set Theory to Solve Classification Problems with Incomplete Data

This paper presents a new approach to fuzzy classification in the case of missing features. The rough set theory is incorporated into neuro-fuzzy structures and the rough-neuro-fuzzy classifier is derived. The architecture of the classifier is determined by the modified indexed center of gravity (MICOG) defuzzification method. The structure of the classifier is presented in a general form, which includes both the Mamdani approach and the logical approach-based on the genuine fuzzy implications. A theorem, which allows the determination of the structures of rough-neuro-fuzzy classifiers based on the MICOG defuzzification, is given and proven. Specific rough-neuro-fuzzy structures based on the Larsen rule, the Reichenbach, and the Kleene-Dienes implications are given in details. In the experiments, it is shown that the classifier with the Dubois-Prade fuzzy implication is characterized by the best performance in the case of missing features.

Rough Sets in the Neuro-Fuzzy Architectures Based on Non-monotonic Fuzzy Implications

In this paper we presented a general solution to compose rough-neuro-fuzzy architectures. Monotonic properties of fuzzy implications were assumed to derive fuzzy systems in the case of missing features. The fuzzy implications satisfying Fodor’s lemma used in logical approach and t-norms used in Mamdani approach are discussed.

On Applying Evolutionary Computation Methods to Optimization of Vacation Cycle Costs in Finite-Buffer Queue

In this paper, problem of positioning and optimization of operation costs for finite-buffer queuing system with exponentially distributed server vacation is investigated. The problem is solved using evolutionary computation methods for independent 2-order hyper exponential input stream of packets and exponential service time distribution. Different scenarios of system operation are analyzed, i.e. different values of parameters of distribution functions describing evolution of the system.

On classification with missing data using rough-neuro-fuzzy systems

On classification with missing data using rough-neuro-fuzzy systems The paper presents a new approach to fuzzy classification in the case of missing data. Rough-fuzzy sets are incorporated into logical type neuro-fuzzy structures and a rough-neuro-fuzzy classifier is derived. Theorems which allow determining the structure of the rough-neuro-fuzzy classifier are given. Several experiments illustrating the performance of the roughneuro-fuzzy classifier working in the case of missing features are described.

Modular type-2 neuro-fuzzy systems

In the paper we study a modular system which can be converted into a type-2 neuro-fuzzy system. The rule base of such system consists of triangular type-2 fuzzy sets. The modular structure is trained using the backpropagation method combined with the AdaBoost algorithm. By applying the type-2 neurofuzzy system, the modular structure is converted into a compressed form. This allows to overcome the training problem of type-2 neuro-fuzzy systems. An illustrative example is given to show the efficiency of our approach in the problems of classification.

Creating learning sets for control systems using an evolutionary method

The acquisition of the knowledge which is useful for developing of artificial intelligence systems is still a problem. We usually ask experts, apply historical data or reap the results of mensuration from a real simulation of the object. In the paper we propose a new algorithm to generate a representative training set. The algorithm is based on analytical or discrete model of the object with applied the k---nn and genetic algorithms. In this paper it is presented the control case of the issue illustrated by well known truck backer---upper problem. The obtained training set can be used for training many AI systems such as neural networks, fuzzy and neuro---fuzzy architectures and k---nn systems.

Neuro-fuzzy Rough Classifier Ensemble

The paper proposes a new ensemble of neuro-fuzzy rough set classifiers. The ensemble uses fuzzy rules derived by the Adaboost metalearning. The rules are used in an ensemble of neuro-fuzzy rough set systems to gain the ability to work with incomplete data (in terms of missing features). This feature is not common among different machine learning methods like neural networks or fuzzy systems. The systems are combined into the larger ensemble to achieve better accuracy. Simulations on a well-known benchmark showed the ability of the proposed system to perform relatively well.

A finite-buffer queue with a single vacation policy

Abstract In this paper, application of an evolutionary strategy to positioning a GI/M/1/N-type finite-buffer queueing system with exhaustive service and a single vacation policy is presented. The examined object is modeled by a conditional joint transform of the first busy period, the first idle time and the number of packets completely served during the first busy period. A mathematical model is defined recursively by means of input distributions. In the paper, an analytical study and numerical experiments are presented. A cost optimization problem is solved using an evolutionary strategy for a class of queueing systems described by exponential and Erlang distributions.