The impact of architecture on the performance of artificial neural networks

A number of researchers have investigated the impact of network architecture on the performance of artificial neural networks. Particular attention has been paid to the impact on the performance of the multi-layer perceptron of architectural issues, and the use of various strategies to attain an optimal network structure. However, there are still perceived limitations with the multi-layer perceptron and networks that employ a different architecture to the multi-layer perceptron have gained in popularity in recent years, particularly, networks that implement a more localised solution, where the solution in one area of the problem space does not impact, or has a minimal impact, on other areas of the space. In this study, we discuss the major architectural issues affecting the performance of a multi-layer perceptron, before moving on to examine in detail the performance of a new localised network, namely the bumptree. The work presented here examines the impact on the performance of artificial neural networks of employing alternative networks to the long established multi-layer perceptron. In particular, networks that impose a solution where the impact of each parameter in the final network architecture has a localised impact on the problem space being modelled are examined. The alternatives examined are the radial basis function and bumptree neural networks, and the impact of architectural issues on the performance of these networks is examined. Particular attention is paid to the bumptree, with new techniques for both developing the bumptree structure and employing this structure to classify patterns being examined.

Parallel strategy for optimal learning in perceptrons

We developed a parallel strategy for learning optimally specific realizable rules by perceptrons, in an online learning scenario. Our result is a generalization of the Caticha–Kinouchi (CK) algorithm developed for learning a perceptron with a synaptic vector drawn from a uniform distribution over the N-dimensional sphere, so called the typical case. Our method outperforms the CK algorithm in almost all possible situations, failing only in a denumerable set of cases. The algorithm is optimal in the sense that it saturates Bayesian bounds when it succeeds.

Multiple conduction paths in boron δ-doped diamond structures

Impedance spectroscopy has been used to investigate conductivity within boron-doped diamond in an intrinsic/delta-doped/intrinsic (i-d-i) multilayer structure. For a 5 nm thick delta layer, three conduction pathways are observed, which can be assigned to transport within the delta layer and to two differing conduction paths in the i-layers adjoining the delta layer. For transport in the i-layers, thermal trapping/detrapping processes can be observed, and only at the highest temperature investigated (673 K) can transport due to a single conduction process be seen. Impedance spectroscopy is an ideal nondestructive tool for investigating the electrical characteristics of complex diamond structures.

Size-dependent radiation tolerance in ion irradiated TiN/AlN nanolayer films

Interface effects on ion-irradiation tolerance properties are investigated in nanolayered TiN/AlN films with individual layer thickness varied from 5 nm to 50 nm, prepared by pulsed laser deposition. Evolution of the microstructure and hardness of the multilayer films are examined on the specimens before and after He ion-implantation to a fluence of 4 × 10 m at 50 keV. The suppression of amorphization in AlN layers and the reduction of radiation-induced softening are observed in all nanolayer films. A clear size-dependent radiation tolerance characteristic is observed in the nanolayer films, i.e., the samples with the optimum layer thickness from 10 nm to 20 nm show the best ion irradiation tolerance properties, and a critical layer thickness of more than 5 nm is necessary to prevent severe intermixing. This study suggests that both the interface characteristics and the critical length scale (layer thickness) contribute to the reduction of the radiation-induced damages in nitride-based ceramic materials. © 2013 Elsevier B.V. All rights reserved.

Replication-based inference algorithms for hard computational problems

Inference algorithms based on evolving interactions between replicated solutions are introduced and analyzed on a prototypical NP-hard problem: the capacity of the binary Ising perceptron. The efficiency of the algorithm is examined numerically against that of the parallel tempering algorithm, showing improved performance in terms of the results obtained, computing requirements and simplicity of implementation. © 2013 American Physical Society.

Nanostructured Cu/Nb multilayers subjected to helium ion-irradiation

Helium ion-irradiation experiments have been performed in single layer Cu films, Nb films and Cu/Nb multilayer films with layer thickness varying from 2.5 nm to 100 nm each layer. Peak helium concentration approaches a few atomic percent with 6-9 displacement-per-atom in Cu and Nb. He bubbles were observed in single layer Cu and Nb films, as well as in Cu 100 nm/Nb 100 nm multilayers with helium bubbles aligned along layer interfaces. Helium bubbles are not resolved via transmission electron microscopy in Cu 2.5 nm/Nb 2.5 nm multilayers. These studies indicate that layer interface may play an important role in annihilating ion-irradiation induced defects such as vacancies and interstitials and have implications in improving the radiation tolerance of metallic materials using nanostructured multilayers. © 2007 Elsevier B.V. All rights reserved.

Strengthening mechanisms in nanostructured copper/304 stainless steel multilayers

Nanostructured Cu/304 stainless steel (SS) multilayers were prepared by magnetron sputtering. 304SS has a face-centered-cubic (fcc) structure in bulk. However, in the Cu/304SS multilayers, the 304SS layers exhibit the fcc structure for layer thickness of =5 nm in epitaxy with the neighboring fcc Cu. For 304SS layer thickness larger than 5 nm, body-centered-cubic (bcc) 304SS grains grow on top of the initial 5 nm fcc SS with the Kurdjumov-Sachs orientation relationship between bcc and fcc SS grains. The maximum hardness of Cu/304SS multilayers is about 5.5 GPa (factor of two enhancement compared to rule-of-mixtures hardness) at a layer thickness of 5 nm. Below 5 nm, hardness decreases with decreasing layer thickness. The peak hardness of fcc/fcc Cu/304SS multilayer is greater than that of Cu/Ni, even though the lattice-parameter mismatch between Cu and Ni is five times greater than that between Cu and 304SS. This result may primarily be attributed to the higher interface barrier stress for single-dislocation transmission across the {111} twinned interfaces in Cu/304SS as compared to the {100} interfaces in Cu/Ni.

Microwave properties of an inhomogeneous optically illuminated plasma in a microstrip gap

The optical illumination of a microstrip gap on a thick semiconductor substrate creates an inhomogeneous electron-hole plasma in the gap region. This allows the study of the propagation mechanism through the plasma region. This paper uses a multilayer plasma model to explain the origin of high losses in such structures. Measured results are shown up to 50 GHz and show good agreement with the simulated multilayer model. The model also allows the estimation of certain key parameters of the plasma, such as carrier density and diffusion length, which are difficult to measure by direct means. The detailed model validation performed here will enable the design of more complex microwave structures based on this architecture. While this paper focuses on monocrystalline silicon as the substrate, the model is easily adaptable to other semiconductor materials such as GaAs.

Compression by replication

A recently introduced inference method based on system replication and an online message passing algorithm is employed to complete a previously suggested compression scheme based on a nonlinear perceptron. The algorithm is shown to approach the information theoretical bounds for compression as the number of replicated systems increases, offering superior performance compared to basic message passing algorithms. In addition, the suggested method does not require fine-tuning of parameters or other complementing heuristic techniques, such as the introduction of inertia terms, to improve convergence rates to nontrivial results. © 2014 American Physical Society.

Linear Classifiers Based on Binary Distributed Representations

Binary distributed representations of vector data (numerical, textual, visual) are investigated in classification tasks. A comparative analysis of results for various methods and tasks using artificial and real-world data is given.

Fourier Neural Networks: An Approach with Sinusoidal Activation Functions

* Supported by INTAS 2000-626, INTAS YSF 03-55-1969, INTAS INNO 182, and TIC 2003-09319-c03-03.

Intellect Sensing of Neural Network that Trained to Classify Complex Signals

An experimental comparison of information features used by neural network is performed. The sensing method was used. Suboptimal classifier agreeable to the gaussian model of the training data was used as a probe. Neural nets with architectures of perceptron and feedforward net with one hidden layer were used. The experiments were carried out with spatial ultrasonic data, which are used for car’s passenger safety system neural controller learning. In this paper we show that a neural network doesn’t fully make use of gaussian components, which are first two moment coefficients of probability distribution. On the contrary, the network can find more complicated regularities inside data vectors and thus shows better results than suboptimal classifier. The parallel connection of suboptimal classifier improves work of modular neural network whereas its connection to the network input improves the specialization effect during training.

MPGN – An Approach for Discovering Class Association Rules

his article presents some of the results of the Ph.D. thesis Class Association Rule Mining Using MultiDimensional Numbered Information Spaces by Iliya Mitov (Institute of Mathematics and Informatics, BAS), successfully defended at Hasselt University, Faculty of Science on 15 November 2011 in Belgium

Automated Assessment Through Integration of Heterogeneous Systems with a Workflow Engine

Report published in the Proceedings of the National Conference on "Education and Research in the Information Society", Plovdiv, May, 2015

Coke formation in the oxidative dehydrogenation of ethylbenzene to styrene by TEOM

A packed bed microbalance reactor setup (TEOM-GC) is used to investigate the formation of coke as a function of time-on-stream on γ-Al2O3 and 3P/SiO2 catalyst samples under different conditions for the ODH reaction of ethylbenzene to styrene. All samples show a linear correlation of the styrene selectivity and yield with the initial coverage of coke. The COX production increases with the coverage of coke. On the 3 wt% P/SiO2 sample, the initial coke build-up is slow and the coke deposition rate increases with time. On alumina-based catalyst samples, a fast initial coke build-up takes place, decreasing with time-on-stream, but the amount of coke does not stabilize. A higher O2 : EB feed ratio results in more coke, and a higher temperature results in less coke. This coking behaviour of Al2O3 can be described by existing "monolayer-multilayer" models. Further, the coverage of coke on the catalyst varies with the position in the bed. For maximal styrene selectivity, the optimal coverage of coke should be sufficient to convert all O2, but as low as possible to prevent selectivity loss by COX production. This is in favour of high temperature and low O2 : EB feed ratios. The optimal coke coverage depends in a complex way on all the parameters: temperature, the O2 : EB feed ratio, reactant concentrations, and the type of starting material. This journal is