Wyniki wyszukiwania - BazTech

1

Self-Correcting Neural Network for Stereo-matching Problem Solving

Laskowski L., Jelonkiewicz J.

Fundamenta Informaticae

|

2015

|

Vol. 138, nr 4

457--482

EN

In the present paper we describe innovative architecture of artificial neural network based on Hopfield structure - Self Correcting Neural Network (SCNN). It is implementation similar to dual mode Hopfield-like network for solving stereo matching problem. Considered network consists of basic layer of neurons implemented as analogue Hopfield-like network and supervising layer. Thanks to the supervising layer, there is a possibility of modification of the connection weights between the neurons in the basic layer. This enables the improvement of the network performance (accuracy). Authors propose a depth map use for image segmentation and objects auto-selection. High enough accuracy of these processes can be achieved when proposed network (SCNN) is applied. Similar idea can be applied also for images noise removal. In the present article we also describe in detail neurons dynamics in the basic and supervising layers of the SCNN. The network considered here was a subject of experimental tests using real stereo pictures as well as simulated stereo images. This enabled calculation of error and direct comparison with classic analogue Hopfield neural network.

2

Multi-stability of in-phase and anti-phase activity patterns in neural networks with inhibitory and electrical synapses

Ostaszewski H., Meyrand P., Branchereau P., Hallam J., Bem T.

Biocybernetics and Biomedical Engineering

|

2010

|

Vol. 30, no. 3

3-17

EN

As shown in modeling and experimental studies, network comprised of spiking cells interconnected by inhibitory and electrical synapses may express different activity patterns without any change of the network topology or parameters. In this study we confirm robust-ness of this phenomenon by demonstrating multi-stability of hybrid networks consisting of biological neurons of different types. Moreover we show here, using relaxation oscillator model cells, that multi-stability of in-phase (IP) and anti-phase (AP) patterns may be expressed in a network fully connected by instantaneous synaptic inhibition and electrical coupling independently of the network size. In such a network a stimulus of a given profile, consisting of depolarizing and hyperpolarizing signals sent to different subpopulations of cells, can evoke direct switching between IP and AP patterns. We also show that similar phenomenon occurs in more realistic network models with sparse connectivity. Our results suggest that transient signals if arriving in a proper time window may instantaneously reconfigure a given spatio-temporal activity pattern expressed by the network into another stable pattern without any change of the network properties.

3

Programowanie analogowych sieci neuronowych

Gola A., Pisarek J.

Prace Naukowe Akademii im. Jana Długosza w Częstochowie. Edukacja Techniczna i Informatyczna

|

2008

|

T. 3

143-153

EN

The project of the analog neural network, constructed on generally accessible electronic components is presented. A digitally programmable potentiometers and operating-amplifiers are used in the construction. The projected by authors computer programs to the coaching of this network are presented too. There are used two algorithms. The linear programming algorithm with the regularization module is applied to the coaching of the lineal network. The gradient algorithm is proposed to the programming of the analog network with any architecture.

4

Crosslinking of new elastomers functionalized with carboxyl groups

Smejda-Krzewicka A., Rzymski W. M.

Polimery

|

2006

|

T. 51, nr 1

66-68

EN

The new carboxylated acrylonitrile-butadiene (XNBR) and styrene-butadiene (XSBR) rubbers produced were crosslinked with zinc oxide. The cure products are characterized by acceptable stress-strain properties. The increase in ZnO amount used, after exceeding of stoichiometric quantity with respect to -COOH groups content, does not improve the stress-strain properties. The curing of new XNBR and XSBR with ZnO leads to the formation of ionic crosslinks. An increase in swelling degree in toluene after action of NH3 vapors, capable to destroy the ionic bonds, confirms this conclusion. The best stress-strain properties of cured products were achieved when carboxylated rubbers were hybrid cured, e.g. using simultaneously zinc oxide and sulfur, in the presence of N-cyclohexyl-2-benzothiazolesulfenamide as vulcanization accelerator. The products of such crosslinking contain both ionic and sulfide bonds and are characterized by the tensile strength of 8 MPa in cured state.

PL

Nowe karboksylowane kauczuki butadienowo-akrylonitrylowe (XNBR) i butadienowo-styrenowe (XSBR) (tabela 1) sieciowano za pomocą tlenku cynku użytego w ilości stechiometrycznej względem grup karboksylowych. Usieciowane elastomery charakteryzują się lepszą wytrzymałością na rozciąganie niż ich nieusieciowane analogi (rys. 1). Zwiększenie dodatku ZnO powyżej stechiometrycznej względem ilości grup -COOH nie wpływa na poprawę właściwości mechanicznych usieciowanych produktów. W wyniku sieciowania XNBR lub XSBR za pomocą ZnO powstają jonowe węzły sieci, rozkładające się pod wpływem par amoniaku (rys. 2). Najlepsze właściwości osiągane są jednak podczas sieciowania hybrydowego, tj. jednocześnie tlenkiem cynku i siarką, w obecności przyspieszacza wulkanizacji. Wytworzone w ten sposób produkty charakteryzują się najlepszą wytrzymałością na rozciąganie (ok. 8 MPa) wynikającą z obecności zarówno jonowych, jak i siarczkowych wiązań poprzecznych.