Parallel-hierarchical Network Research Articles

NEUROBIOLOGICAL PROPERTIES OF THE STRUCTURE OF THE PARALLEL-HIERARCHICAL NETWORK AND ITS USAGE FOR PATTERN RECOGNITION

The paper presents the analysis of neurobiological data on the existence of the structure of a parallel-hierarchical network. Discussed method of parallel-hierarchical transformation based on population coding and its application for the pattern recognition task. Based on the analysis, we can conclude that using the methods proposed, it is possible to measure the geometric parameters and properties of images, which can significantly increase the efficiency of processing, in particular estimating the center of mass based on moment characteristics. Experimental results demonstrate that due to various destabilizing factors, accurately measuring the energy center coordinates of laser beam spot images is challenging. However, training the PI network and classifying the fragments into "good" and "bad" can considerably enhance the accuracy of these measurements.

Analysis of the main provisions of the theory of parallel-hierarchical transformations

The article presents an analysis of the main principles of parallel-hierarchical transformations theory. The continuous movement of society towards the automation of everyday life requires the creation of fundamentally new software and hardware solutions. Considering the current physical limitations of integrated circuits, it is evident that improving software processing is the way to go. The main problem lies in the increasing complexity of architecture and supporting such code. The ideas of parallel-hierarchical networks allow for a significant increase in processing speed through process parallelization while maintaining the relative simplicity of the software solution's architecture. The proposed structure of the parallel-hierarchical network allows for modelling the operation principle of a distributed neural network and forms a deterministic network using spatial-temporal division. The general rules of direct and inverse parallel-hierarchical transformation and their application to image recognition tasks are discussed. A block diagram of the algorithm for the basic model of nonlinear direct network transformation is shown. A mathematical model of direct parallel-hierarchical transformation is presented using an example. Compared to known numerical transformation methods involving simple operations like addition, the model enables complex functional signal processing in real-time scale, as well as unambiguity and reversibility with good convergence of the computational process.

DEVELOPMENT OF MATHEMATICAL MODELS OF PARALLELHIERARCHICAL TRANSFORMATION FOR EFFICIENT IMAGE PROCESSING

This article considers the problems of modern models and methods of recognition of information and images, the analogy of learning parallel-hierarchical networks with RBF networks. During the analysis, it was determined that image processing using the mask method does not use all the power of multi-core systems. The analysis of parallel-hierarchical transformation and image processing is carried out. The analysis of the mathematical model of parallel-hierarchical transformation for speed increase by means of parallel transformation is carried out. The algorithm was simulated using the C # language, and artificially created images using microwave were used for the test. The simulation was performed for complete PI conversion, the simulation results are shown in the tables. Parallel-hierarchical transformation for image processing has great potential in transport systems, it can be used in tasks where it is necessary to quickly analyze images from the camera, such as the presence of obstacles in the trajectory of traffic and collision avoidance, analysis of asphalt damage on the trajectory and suspension adjustment to level the effect. The results obtained in the modeling process allow us to conclude that the speed of parallel-hierarchical transformation in image processing problems.

Extracting Parallel Sentences from Nonparallel Corpora Using Parallel Hierarchical Attention Network.

Collecting parallel sentences from nonparallel data is a long-standing natural language processing research problem. In particular, parallel training sentences are very important for the quality of machine translation systems. While many existing methods have shown encouraging results, they cannot learn various alignment weights in parallel sentences. To address this issue, we propose a novel parallel hierarchical attention neural network which encodes monolingual sentences versus bilingual sentences and construct a classifier to extract parallel sentences. In particular, our attention mechanism structure can learn different alignment weights of words in parallel sentences. Experimental results show that our model can obtain state-of-the-art performance on the English-French, English-German, and English-Chinese dataset of BUCC 2017 shared task about parallel sentences' extraction.

Development of a method of processing images of laser beam bands with the use of parallel­hierarchic networks

A method for processing images of spots of laser beams is presented for systems using a modern and promising form of communication using atmospheric-optical communication lines (AOCL). The method allows to significantly expand the use of laser technology in information transfer systems by increasing speed, their efficiency and the characteristics of the positioning subsystem in them. This is important because under the influence of the atmosphere (rain, fog, dust, snow) the laser beam is deformed, and this, in turn, affects the quality of the results obtained from such systems. The necessity of filtering images received from the transmitter by using adaptive methods of processing information in parallel hierarchical networks is substantiated. In order to accelerate the processing of images of spots of laser beams, a specialized random access memory (RAM) is developed, which is used in modeling the developed method for processing images of spots of laser beams. Based on the obtained simulation data, the results of predicting the location of the receiver and the system performance obtained with the presented method are 15–20 % better than the results obtained using known methods. As a result of the development of the presented method, it is possible to reduce the number of accesses to RAM by 2 times, and therefore reduce the load on the memory. The operation of reading 2-port memory has been improved, it allows to create a displacement vector of parallel-hierarchical transformation in one cycle, dividing the width of the record and reading the data bus, which in turn increases the RAM efficiency

Информационная технология классификации изображений пятен лазерных пучков и прогнозирования координат их энергетических центров с использованием параллельно-иерархической сети

Информационная технология классификации изображений пятен лазерных пучков и прогнозирования координат их энергетических центров с использованием параллельно-иерархической сети

Laser beam image classification methods with the use of parallel-hierarchical networks running on a programmable logic device

We examine the methods of laser beam classification and their uses. We discuss the necessity of noise filters using adaptive methods in beam reflection, such as parallel-hierarchical networks. A demonstration of this network is shown on a programmable logic device.

Оброблення зображень плям лазерних пучків із застосуванням паралельно-ієрархічних мереж

The basic steps for classification and prediction coordinates energy centers image spots of the laser beam, which makes it possible to develop a new intelligent technologies for classification and prediction of the position coordinate their energy centers. Experiments on the comparative evaluation of prediction based on known neural networks, and proposed method using a parallel-hierarchical network was conducted.

Method of predicting the position of the energy center of the image of a laser beam using a parallel–hierarchical network

This paper presents a prediction method using a parallel---hierarchical (PH) network and hyperbolic smoothing of empirical data. The average prediction error is 0.55% for the developed method and 1.62% for neural networks; therefore, this method is more efficient as applied to real-time systems than traditional neural networks due to the use of the PH network and hyperbolic smoothing in implementing the operation of predicting the positions of energy centers of laser beam spot images for optical communication systems.

Method of forecasting energy center positions of laser beam spot images using a parallel hierarchical network for optical communication systems

A forecasting method, based on the parallel-hierarchical (PH) network and hyperbolic smoothing of empirical data, is presented in this paper. Preceding values of the time series, hyperbolic smoothing, and PH network data are used for forecasting. To determine a position of the next route fragment in relation to X and Y axes, hyperbola parameters are sent to the route parameter forecasting system. In the results synchronization block, network-processed data arrive to the database where a sample of most correlated data is drawn using service parameters of the PH network. An average prediction error is 0.55% for the developed method and 1.62% for neural networks. That is why, due to the use of the PH network and hyperbolic smoothing, the developed method is more efficient for real-time systems than traditional neural networks in forecasting energy center positions of laser beam spot images for optical communication systems.

Modeling of a method of parallel hierarchical transformation for fast recognition of dynamic images

Principles necessary to develop a method and computational facilities for the parallel hierarchical transformation based on high-performance GPUs are discussed in the paper. Mathematic models of the parallel hierarchical (PH) network training for the transformation and a PH network training method for recognition of dynamic images are developed.

Method of Parallel-Hierarchical Network Self-Training and its Application for Pattern Classification and Recognition

Method of Parallel-Hierarchical Network Self-Training and its Application for Pattern Classification and Recognition

Parallel-hierarchical network learning methods and their application to pattern recognition

This paper considers provisions necessary for developing parallel-hierarchical network learning methods underlain by the idea of population coding in an artificial neural network and its approximation to natural neural networks. Mathematical parallel-hierarchical network learning models and a combined parallel-hierarchical network learning method are developed for recognizing static and dynamic patterns.

A method of organization of a parallel-hierarchical network for image recognition

An explicit formula of a multivariate B-spline, some useful investigations in the field of linear transformations of independent exponentially distributed random variables, representation of theirs density functions with the help of multivariate exponential spline functions, and their usage are considered. The consideration is illustrated by an appropriate example.

Methodological Principles of Pyramidal and Parallel-Hierarchical Image Processing on the Base of Neural-Like Network Systems

Abstract —The scientific research deals with organization principles of parallel-hierarchical transform for multistage perception and processing, compression and recognition of information in informational structure and computational systems which make use of computational scheme similar to neural. Unified methodological approach was developed for analysis of parallel processes. This approach considers influence of structural hierarchy in dynamics, in other words it tracks processes of spatial areas transformation of correlated and generation of uncorrelated in time elements of generated network, at the time of transition of the network from one stable stage to another. Index Terms—parallel processing, parallel-hierarchical network, image processing, pattern recognition I. INTRODUCTION Constantly growing volume of data and computations, needed for processing of large arrays of information, for instance, images, requires higher level of performance of the systems involved in this process. Since the density of elements “package” in integrated circuits is determined by physical constraints, operation speed is limited by final speed of electromagnetic modes propagation from one element to another. It is possible to overcome this physical barrier only by means of pipelining of computational processes occurring in the system, that, in its turn, results in complication of system architecture. Intelligent processing of information requires consideration of each element in a certain context of its connections, and it is possible only in computing system, having topographic structure with 3D location of processing elements (PE). Such a structure of the system perfectly matches natural neurolike perception of sensor information [1, 2, 3]. The idea of hierarchical pyramidal processing of information was suggested for the first time in [4, 5, 6, 7]. Highly-efficient system of information processing could be realized adapting architecture to the corresponding data structure. But structure of data in the process of pyramidal processing changes from large fixed array on lower level to small flexible structure on upper level. The most interesting case is uniform non-distributed computational structures, which correspond to SIMD-syste m class, where some levels of identical PE operate in SIMD mode. Each level contains a great number of simple PE. Su ch system as PCLIP, PAPIA, GAM, SPHINX [8] belong to this group, as well as structures suggested by the authors and protected by certificates and patent for a device intended for realization of pyramidal processing of information [5]. In more sophisticated case of uniform distributed pyramidal computational structures some powerful identical processing units are integrated into hierarchical pyramidal structure. Each processing unit corresponds to a part of processed data. Such pyramidal system can op erate both in SIMD and in MIMD-modes. Computational structures [1, 5] suggested by the author and protected by th e certificates for a means and device, as well as systems Uhrs, Array/Net and EGPA [8] belong to this family. II. THE MAIN PRINCIPLES OF PYRAMIDAL AND PARALLEL-HIERARCHICAL PROCESSING OF DIGITAL INFORMATION Principle of construction of pyramidal hierarchical structure of data can be defined as a sequence of data arrays of the same information field at different levels of division: P = (A0, A1, A2, , AL), where A

An approach to restoration and recovery problems using parallel hierarchical neural networks

In fMRI data analysis it has been shown that for a wide range of situations the hemodynamic response function (HRF) can be reasonably characterized as the impulse response function of a linear and time invariant system. An accurate and robust extraction of the HRF is essential to infer quantitative information about the relative timing of the neuronal events in different brain regions. When no assumptions are made about the HRF shape, it is most commonly estimated using time windowed averaging or a least squares estimated general linear model based on either Fourier or delta basis functions. Recently, regularization methods have been employed to increase the estimation efficiency of the HRF; typically these methods produce more accurate HRF estimates than the least squares approach [Goutte, C., Nielsen, F.A., Hansen, L.K., 2000. Modeling the Haemodynamic Response in fMRI Using Smooth FIR Filters. IEEE Trans. Med. Imag. 19(12), 1188–1201.]. Here, we use simulations to clarify the relative merit of temporal regularization based methods compared to the least squares methods with respect to the accuracy of estimating certain characteristics of the HRF such as time to peak (TTP), height (HR) and width (W) of the response. We implemented a Bayesian approach proposed by Marrelec et al. [Marrelec, G., Benali, H., Ciuciu, P., Pelegrini-Issac, M., Poline, J.-B., 2003. Robust Estimation of the Hemodynamic Response Function in Event-Related BOLD fMRI Using Basic Physiological Information. Hum. Brain Mapp. 19, 1–17., Marrelec, G., Benali, H., Ciuciu, P., Poline, J.B. Bayesian estimation of the hemodynamic of the hemodynamic response function in functional MRI. In: R. F, editor; 2001; Melville. p 229–247.] and its deterministic counterpart based on a combination of Tikhonov regularization [Tikhonov, A.N., Arsenin, V.Y., 1977. Solution of ill-posed problems. Washington DC: W.H. Winston.] and generalized cross-validation (GCV) [Wahba, G., 1990. Spline Models for Observational Data. Philadelphia: SIAM.] for selecting the regularization parameter. The performance of both methods is compared with least square estimates as a function of temporal resolution, color and strength of the noise, and the type of stimulus sequences used. In almost all situations, under the considered assumptions (e.g. linearity, time invariance and smooth HRF), the regularization-based techniques more accurately characterize the HRF compared to the least-squares method. Our results clarify the effects of temporal resolution, noise color, and experimental design on the accuracy of HRF estimation.