I. G. Persiantsev

Multifold Acceleration of Neural Network Computations Using GPU

With emergence of graphics processing units (GPU) of the latest generation, it became possible to undertake neural network based computations using GPU on serially produced video display adapters. In this study, NVIDIA CUDA technology has been used to implement standard back-propagation algorithm for training multiple perceptrons simultaneously on GPU. For the problem considered, GPU-based implementation (on NVIDIA GTX 260 GPU) has lead to a 50x speed increase compared to a highly optimized CPU-based computer program, and more than 150x compared to a commercially available CPU-based software (NeuroShell 2) (AMD Athlon 64 Dual core 6000+ processor).

Application of artificial neural networks to solve problems of identification and determination of concentration of salts in multi-component water solutions by Raman spectra

In this paper, the results of elaboration and comparative analysis of approaches concerned with application of neural network algorithms for effective solution of problem of pattern recognition (inverse problem with discrete output) along with inverse problem with continuous output are presented. Consideration is carried out at the example of problem of identification and determination of concentrations of inorganic salts in multi-component water solutions by Raman spectra. The studied approach is concerned with solution of both problems (classification and determination of concentrations) using a single neural network trained on experimental or quasi-model data.

ONE-PARAMETER REPRESENTATION OF THE DAILY AVERAGED SOLAR-WIND VELOCITY

An empirical formula was found to describe the dependence V(S) of the daily average solar-wind velocity V on the coronal-hole area S on the visible side of the Sun in the form of first-and second-order Taylor expansions. The results can be used for approximate evaluation of the solar-wind velocity at the Earth’s orbit from coronal-hole observations.

Adaptive methods for solving inverse problems in laser raman spectroscopy of multi-component solutions

This study provides comparative analysis of approaches connected with application of neural network based algorithms for efficient solution of pattern recognition problem (inverse problem with discrete output) combined with solution of inverse problem with continuous output. The analysis is performed at the example of the problem of identification and determination of concentrations of inorganic salts in multi-component aqueous solutions by Raman spectrum.

Comparison of the quality of solving the inverse problems of spectroscopy of multi-component solutions with neural network methods and with the method of projection to latent structures

This study provides comparative analysis of application of artificial neural networks and method of projection to latent structures (partial least squares) for simultaneous determination of types and concentrations of dissolved inorganic salts in multicomponent water solutions by Raman spectra. It is shown that the method of projection to latent structures has several advantages, such as the quality of the solution and the time of construction of a regression model, when solving problems with low level of nonlinearity.

Neural Network Approaches to Solution of the Inverse Problem of Identification and Determination of Partial Concentrations of Salts in Multi-сomponent Water Solutions

The studied inverse problem is determination of partial concentrations of inorganic salts in multi-component water solutions by their Raman spectra. The problem is naturally divided into two parts: 1) determination of the component composition of the solution, i.e. which salts are present and which not; 2) determination of the partial concentration of each of the salts present in the solution. Within the first approach, both parts of the problem are solved simultaneously, with a single neural network (perceptron) with several outputs, each of them estimating the concentration of the corresponding salt. The second approach uses data clusterization by Kohonen networks for consequent identification of component composition of the solution by the cluster, which the spectrum of this solution falls into. Both approaches and their results are discussed in this paper.

Application of neural networks to fluorescent diagnostics of organic pollution in water

Rapid diagnosis of pollution is one of the key tasks in the field of ecological monitoring of natural and technogeneous environment. One of the promising methods of fluorescent diagnosis of organic pollution of water environment is the registration and analysis of 2D spectral fluorescent signatures (SFS). The neural networks-based system suggested in this paper is intended for solving the problem of detection, identification, and concentration measurement of water environmental pollution. The suggested system uses SFS as input pattern and allows one to build a rapid diagnosis system for ecological monitoring.

Comparison of Adaptive Algorithms for Significant Feature Selection in Neural Network Based Solution of the Inverse Problem of Electrical Prospecting

One of the important directions of research in geophysical electrical prospecting is solution of inverse problems (IP), in particular, the IP of magnetotellurics --- the problem of determining the distribution of electrical conductivity in the thickness of earth by the values of electromagnetic field induced by ionosphere sources, observed on earth surface. Solution of this IP is hampered by very high dimensionality of the input data (~103---104). Selection of the most significant features for each determined parameter makes it possible to simplify the IP and to increase the precision of its solution. This paper presents a comparison of two modifications of the developed algorithm for multi-step selection of significant features and the results of their application.

Neural Network Approaches to Solution of the Inverse Problem of Identification and Determination of the Ionic Composition of Multi-component Water Solutions

The studied inverse problem is determination of ionic composition of inorganic salts (concentrations of up to 10 ions) in multi-component water solutions by their Raman spectra. The regression problem was solved in two ways: 1) by a multilayer perceptron trained on the large dataset, composed of spectra of all possible mixing options of ions in water; 2) dividing the data set into compact clusters and creating regression models for each cluster separately. Within the first approach, we used supervised training of neural network, achieving good results. Unfortunately, this method isn’t stable enough; the results depend on data subdivision into training, test, and out-of-sample sets. In the second approach, we used algorithms of unsupervised learning for data clustering: Kohonen networks, k-means, k-medoids and hierarchical clustering, and built partial least squares regression models on the small datasets of each cluster. Both approaches and their results are discussed in this paper.

Study of Influence of Parameter Grouping on the Error of Neural Network Solution of the Inverse Problem of Electrical Prospecting

In the electrical prospecting inverse problem, the sought-for distribution of electrical conductivity in Earth stratum is described by dividing the studied section into blocks arranged in layers, with determination of electrical conductivity in the center of each block. This inverse problem can be solved separately for each block, or simultaneously for a group of blocks. In this study, the dependence of solution error on the number of blocks for simultaneous solution of the problem with a single neural network, and on the method of their choice, was investigated.