Wyniki wyszukiwania - BazTech

1

Basic concepts of dynamic recurrent neural networks development

Boyko N., Pobereyko P.

ECONTECHMOD : An International Quarterly Journal on Economics of Technology and Modelling Processes

|

2016

|

Vol. 5, No 2

63--68

EN

In this work formulated relevance, set out an analytical review of existing approaches to the research recurrent neural networks (RNN) and defined precondition appearance a new direction in the field neuroinformatics – reservoir computing. Shows generalized classification neural network (NN) and briefly described main types dynamics and modes RNN. Described topology, structure and features of the model NN with different nonlinear functions and with possible areas of progress. Characterized and systematized wellknown learning methods RNN and conducted their classification by categories. Determined the place RNN with unsteady dynamics of other classes RNN. Deals with the main parameters and terminology, which used to describe models RNN. Briefly described practical implementation recurrent neural networks in different areas natural sciences and humanities, and outlines and systematized main deficiencies and the advantages of using different RNN. The systematization of known recurrent neural networks and methods of their study is performed and on this basis the generalized classification of neural networks was proposed.

2

The UD RLS Algorithm for Training Feedforward Neural Networks

Bilski J.

International Journal of Applied Mathematics and Computer Science

|

2005

|

Vol. 15, no 1

115-123

EN

A new algorithm for training feedforward multilayer neural networks is proposed. It is based on recursive least squares procedures and U-D factorization, which is a well-known technique in filter theory. It will be shown that due to the U-D factorization method, our algorithm requires fewer computations than the classical RLS applied to feedforward multilayer neural network training.

3

Beta neuro-fuzzy systems

Alimi A. M.

TASK Quarterly : scientific bulletin of Academic Computer Centre in Gdansk

|

2003

|

Vol. 7, No 1

23-41

EN

In this paper we present the Beta function and its main properties. A key feature of the Beta function, which is given by the central-limit theorem, is also given. We then introduce a new category of neural networks based on a new kernel: the Beta function. Next, we investigate the use of Beta fuzzy basis functions for the design of fuzzy logic systems. The functional equivalence between Beta-based function neural networks and Beta fuzzy logic systems is then shown with the introduction of Beta neuro-fuzzy systems. By using the SW theorem and expanding the output of the Beta neuro-fuzzy system into a series of Beta fuzzy-based functions, we prove that one can uniformly approximate any real continuous function on a compact set to any arbitrary accuracy. Finally, a learning algorithm of the Beta neuro-fuzzy system is described and illustrated with numerical examples.

4

Neuro-fuzzy TSK network for approximation of static and dynamic functions

Linh T., Osowski S.

Control and Cybernetics

|

2002

|

Vol. 31, no 2

309-326

EN

The paper presents the neuro-fuzzy network in application to the approximation of the static and dynamic functions. The network implements the Takagi-Sugeno inference rules. The learning algorithm is based on the hybrid approach, splitting the learning phase into two stages : the adaptation of the linear output weights using the SVD algorithm and the conventional steepest descent backpropagation rule in application to the adaptation of the nonlinear parameters of the membership functions. The new approach to the generation of the inference rules, based on the fuzzy self-organization is proposed and the algorithm of automatic determination of the number of these rules has been also implemented. The method has been applied for the off-line modelling of static nonlinear relations and on-line simulation of the dynamic systems.

5

Diagnosis System Based on Multiple Neural Classifiers

Marciniak A., Korbicz J.

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2001

|

vol. 49, nr 4

681-701

EN

A new methodology for improving the performance and training of neural network classifiers employed in diagnostics is presented. The main idea is based on using redundant classifiers in an ensemble in order to guarantee the best generalisation ability of the diagnosis system. A brief survey of some commonly used methods for combining outputs in the ensemble is made. As compared to previous designs, a novel method for output combination is introduced. The proposed technique consist in considering the classes independently of one another and calculating the importance parameters, i.e. the weights, for individual outputs of the networks. In order to draw a comparison with previous methods, a real data medical benchmark is used. To improve the results of the ensemble, Negative Correlation Learning was applied.

6

Dynamic Neural Networks for Process Modelling in Fault Detection and Isolation Systems

Korbicz J., Patan K., Obuchowicz A.

International Journal of Applied Mathematics and Computer Science

|

1999

|

Vol. 9, no 3

519-546

EN

A fault diagnosis scheme for unknown nonlinear dynamic systems with modules of residual generation and residual evaluation is considered. Main emphasis is placed upon designing a bank of neural networks with dynamic neurons that model a system diagnosed at normal and faulty operating points.To improve the quality of neural modelling, two optimization problems are included in the construction of such dynamic networks: searching for an optimal network architecture and the network training algorithm. To find a good solution, the effective well-known cascade-correlation algorithm is adapted here. The residuals generated by a bank of neural models are then evaluated by means of pattern classification. To illustrate the effectiveness of our approach, two applications are presented: a neural model of Narendra's system and a fault detection and identification system for the two-tank process.