Wyniki wyszukiwania - BazTech

1

Emerging modularity during the evolution of neural networks

Praczyk Tomasz

Journal of Artificial Intelligence and Soft Computing Research

|

2023

|

Vol. 13, No. 2

107--126

EN

Modularity is a feature of most small, medium and large–scale living organisms that has evolved over many years of evolution. A lot of artificial systems are also modular, however, in this case, the modularity is the most frequently a consequence of a handmade design process. Modular systems that emerge automatically, as a result of a learning process, are very rare. What is more, we do not know mechanisms which result in modularity. The main goal of the paper is to continue the work of other researchers on the origins of modularity, which is a form of optimal organization of matter, and the mechanisms that led to the spontaneous formation of modular living forms in the process of evolution in response to limited resources and environmental variability. The paper focuses on artificial neural networks and proposes a number of mechanisms operating at the genetic level, both those borrowed from the natural world and those designed by hand, the use of which may lead to network modularity and hopefully to an increase in their effectiveness. In addition, the influence of external factors on the shape of the networks, such as the variability of tasks and the conditions in which these tasks are performed, is also analyzed. The analysis is performed using the Hill Climb Assembler Encoding constructive neuro-evolutionary algorithm. The algorithm was extended with various module-oriented mechanisms and tested under various conditions. The aim of the tests was to investigate how individual mechanisms involved in the evolutionary process and factors external to this process affect modularity and efficiency of neural networks.

2

Optimization of a modular neural network for pattern recognition using parallel genetic algorithm

Cárdenas M., Melin P., Cruz L.

Journal of Automation Mobile Robotics and Intelligent Systems

|

2011

|

Vol. 5, No. 1

77-84

EN

In this paper, the implementation of a Parallel Genetic Algorithm (PGA) for the training stage, and the optimi zation of a monolithic and modular neural network, for pattern recognition are presented. The optimization con sists in obtaining the best architecture in layers, and neu rons per layer achieving the less training error in a shor ter time. The implementation was performed in a multicore architecture, using parallel programming techniques to exploit its resources. We present the results obtained in terms of performance by comparing results of the training stage for sequential and parallel implementations.

3

Type-1 and Type-2 Fuzzy Inference Systems as Integration Methods in Modular Neural Networks for Multimodal Biometry and its Optimization with Genetic Algorithms

Hidalgo D., Castillo O., Melin P.

Journal of Automation Mobile Robotics and Intelligent Systems

|

2008

|

Vol. 2, No. 1

59-73

EN

We describe in this paper a comparative study between Fuzzy Inference Systems as methods of integration in modular neural networks for multimodal biometry. These methods of integration are based on techniques of type-1 fuzzy logic and type-2 fuzzy logic. Also, the fuzzy systems are optimized with simple genetic algorithms. First, we considered the use of type-1 fuzzy logic and later the approach with type-2 fuzzy logic. The fuzzy systems were developed using genetic algorithms to handle fuzzy inference systems with different membership functions, like the triangular, trapezoidal and Gaussian; since these algorithms generate the fuzzy systems automatically. Then the response integration of the modular neural network was tested with the optimized fuzzy integration systems. The comparative study of type-1 and type-2 fuzzy inference systems was made to observe the behavior of the two different integration methods for modular neural networks for multimodal biometry.

4

Wykorzystanie neuronowych sieci modularnych do diagnostyki analogowych układów elektronicznych

Kurowski I., Rutkowski J.

Elektronika : konstrukcje, technologie, zastosowania

|

2007

|

Vol. 48, nr 11

37-39

PL

Przedmiotem artykułu jest system diagnostyczny służący do lokalizacji pojedynczych uszkodzeń analogowych układów elektronicznych. Na etapie przedtestowym dla diagnozowanego układu elektronicznego skonstruowano słownik uszkodzeń na podstawie symulacji komputerowych dla każdego analizowanego stanu obwodu. Jako pobudzenie wykorzystano zarówno zasilanie układu, jak i podawany na wejście obwodu sygnał sinusoidalny. Na wyjściu badanego obwodu mierzono moduł i fazę napięcia. Zespół dwóch sieci neuronowych o radialnych funkcjach bazowych został użyty jako klasyfikator uszkodzeń (katastroficznych i parametrycznych). Odpowiedź końcową dwóch równolegle połączonych sieci neuronowych wyznaczono z wykorzystaniem przestrzeni BKS (Behavior Knowledge Space).

EN

This paper presents a diagnostic system created for localization of faults in analog electronic circuits. The system is based on dictionary method. As stimu oflation signals, dc supply voltage and sinusoidal input signal have been used. The frequency of sinusoidal signal is variable. Voltages and phases in output of the circuit are measured to create dictionary of faults. Radial Basis Function (RBF) neural networks were used as neural classifiers. Behavior Knowledge Space (BKS) was used to find the optimum final answer of parallel classifiers. Software which has been used to generate teaching-testing files had been created in Borland Delphi. Simulations of electronic circuits have been realized by SPICE. Software, used to simulate neural classifiers has been created in Matlab.

5

Membership function - ARTMAP neural networks

Sinčák P., Hric M., Vaščák J.

TASK Quarterly : scientific bulletin of Academic Computer Centre in Gdansk

|

2003

|

Vol. 7, No 1

43-52

EN

The project deals with the application of computational intelligence (CI) tools for multispectral image classification. Pattern Recognition scheme is a global approach where the classification part is playing an important role to achieve the highest classification accuracy. Multispectral images are data mainly used in remote sensing and this kind of classification is very difficult to assess the accuracy of classification results. There is a feedback problem in adjusting the parts of pattern recognition scheme. Precise classification accuracy assessment is almost impossible to obtain, being an extremely laborious procedure. The paper presents simple neural networks for multispectral image classification, ARTMAP-like neural networks as more sophisticated tools for classification, and a modular approach to achieve the highest classification accuracy of multispectral images. There is a strong link to advances in computer technology, which gives much better conditions for modelling more sophisticated classifiers for multispectral images.

6

Automatic human face recognition using modular neural networks

El-Bakry H. M.

Machine Graphics and Vision

|

2001

|

Vol. 10, No. 1

47-73

EN

In this paper, a fast biometric system for personal identification through face recognition is introduced. In the detection phase, a fast algorithm for face detection is combined with cooperative modular neural networks (MNNs) to enhance the performance of the detection process. A simple design for cooperative modular neural networks is described to solve this problem by dividing the data into three groups. Furthermore, a new faster face detection approach is presented through image decomposition into many sub-images and applying cross correlation in frequency domain between each sub-image and the weights of the hidden layer. For the recognition phase, a new concept for rotation invariant based on Fourier descriptors and neural networks is presented. Although, the magnitude of the Fourier descriptors is translation invariant, there is no need for scaling or translation invariance. This is because the face sub-image (20 x 20 pixels) is segmented from the whole image during the detection process. The feature extraction algorithm based on Fourier descriptors is modified to reduce the number of neurons is the hidden layer. The second stage extracts wavelet coefficients of the resulted Fourier descriptors before application to neural network. The final vector is fed to a neural net for face classification. Moreover, a modified hierarchy soft decision tree of neural networks is introduced for face recignition. Compared with previous results, the proposed algorithm shown good performance on recognizing human faces with glass, bread, rotation, scaling, occlusion, noise, or change in illumination. Also, the response time is reduced.