Kohonen's Feature Map Research Articles

Стохастическая роевая кластеризация в задачах автоматизированной обработки данных, представленных на естественном языке

Consider natural language data processing technology based on non-linear dimensionality reduction method which takes into account the discriminating power of the solution found for given values of the categorical variable associated with each observation. Stochastic optimization method known as the “Particle swarm optimization” is proposed to found characteristics that ensure the best separation of observations in terms of a given quality functional. The basis for evaluating the quality of the solution lies in the purity of the clusters obtained with the k-means method, or with using self-organizing Kohonen feature maps.

Predictive models reduce talent development costs in female gymnastics

ABSTRACTThis retrospective study focuses on the comparison of different predictive models based on the results of a talent identification test battery for female gymnasts. We studied to what extent these models have the potential to optimise selection procedures, and at the same time reduce talent development costs in female artistic gymnastics. The dropout rate of 243 female elite gymnasts was investigated, 5 years past talent selection, using linear (discriminant analysis) and non-linear predictive models (Kohonen feature maps and multilayer perceptron). The coaches classified 51.9% of the participants correct. Discriminant analysis improved the correct classification to 71.6% while the non-linear technique of Kohonen feature maps reached 73.7% correctness. Application of the multilayer perceptron even classified 79.8% of the gymnasts correctly. The combination of different predictive models for talent selection can avoid deselection of high-potential female gymnasts. The selection procedure based upon the different statistical analyses results in decrease of 33.3% of cost because the pool of selected athletes can be reduced to 92 instead of 138 gymnasts (as selected by the coaches). Reduction of the costs allows the limited resources to be fully invested in the high-potential athletes.

Variable-sized Fast Kohonen Feature Map Associative Memory using Area Representation for Sequential Patterns—Association Ability for Sequential Patterns including Common Terms—

Variable-sized Fast Kohonen Feature Map Associative Memory using Area Representation for Sequential Patterns—Association Ability for Sequential Patterns including Common Terms—

Neural Network Modelling and Dynamical System Theory

Recent studies have explored the organization of player movements in team sports using a range of statistical tools. However, the factors that best explain the performance of association football teams remain elusive. Arguably, this is due to the high-dimensional behavioural outputs that illustrate the complex, evolving configurations typical of team games. According to dynamical system analysts, movement patterns in team sports exhibit nonlinear self-organizing features. Nonlinear processing tools (i.e. Artificial Neural Networks; ANNs) are becoming increasingly popular to investigate the coordination of participants in sports competitions. ANNs are well suited to describing high-dimensional data sets with nonlinear attributes, however, limited information concerning the processes required to apply ANNs exists. This review investigates the relative value of various ANN learning approaches used in sports performance analysis of team sports focusing on potential applications for association football. Sixty-two research sources were summarized and reviewed from electronic literature search engines such as SPORTDiscus, Google Scholar, IEEE Xplore, Scirus, ScienceDirect and Elsevier. Typical ANN learning algorithms can be adapted to perform pattern recognition and pattern classification. Particularly, dimensionality reduction by a Kohonen feature map (KFM) can compress chaotic high-dimensional datasets into low-dimensional relevant information. Such information would be useful for developing effective training drills that should enhance self-organizing coordination among players. We conclude that ANN-based qualitative analysis is a promising approach to understand the dynamical attributes of association football players.

Speedup of color palette indexing in self-organization of Kohonen feature map

Based on the self-organization of Kohonen feature map (SOFM), recently, Pei et al. presented an efficient color palette indexing method to construct a color table for compression. Taking the palette indexing method as a representative, this paper presents two new strategies, the pruning-based search strategy and the lookup table (LUT)-based update strategy, to speed up the learning process in the SOFM. Based on four typical testing images, experimental results illustrate that our proposed two strategies have 35% execution-time improvement ratio in average. The practical improvement ratio is very close to that in the theoretical analysis.

Optimization of image processing techniques using neural networks

This paper reviews the application of artificial neural networks in image preprocessing. Neural networks, especially uses feed-forward neural networks, Kohonen feature maps, back-propagation neural...

Applications of neural networks in training science

Training science views itself as an integrated and applied science, developing practical measures founded on scientific method. Therefore, it demands consideration of a wide spectrum of approaches and methods. Especially in the field of competitive sports, research questions are usually located in complex environments, so that mainly field studies are drawn upon to obtain broad external validity. Here, the interrelations between different variables or variable sets are mostly of a nonlinear character. In these cases, methods like neural networks, e.g., the pattern recognizing methods of Self-Organizing Kohonen Feature Maps or similar instruments to identify interactions might be successfully applied to analyze data. Following on from a classification of data analysis methods in training-science research, the aim of the contribution is to give examples of varied sports in which network approaches can be effectually used in training science. First, two examples are given in which neural networks are employed for pattern recognition. While one investigation deals with the detection of sporting talent in swimming, the other is located in game sports research, identifying tactical patterns in team handball. The third and last example shows how an artificial neural network can be used to predict competitive performance in swimming.

Net-based phase-analysis in motion processes

Back in the 1990s, one of the major problems when analysing motion processes often was a lack of data. In recent times, the situation has completely changed. Data are available nearly unlimitedly, and the problem now is to detect the important information hidden in that huge amount of automatically recorded data. Data-mining approaches are often not really helpful as long as it is not clear what to look for or what the striking features are. During the last 10 years, artificial neural networks of type Kohonen Feature Map (KFM) became more and more helpful in the area of motion data analysis by reducing data and classifying them to useful information [see W. Schöllhorn and J. Perl, Prozessanalysen in der Bewegungs- und Sportspielforschung, Spectrum der Sportwissenschaft 14 (1) (2002), pp. 30–52 and J. Perl, A neural network approach to movement pattern analysis. Hum. Mov. Sci. 23 (2004), pp. 605–620]. It should be added that for some of the described applications, in particular in the case of two-level analysis in Section 2.3, the special KFM-derivate Dynamically Controlled Network (DyCoN) is necessary [see J. Perl, DyCoN: Ein neuer Ansatz zur Modellierung und Analyse von Sportspie-Prozessen mit Hilfe neuronaler Netze, in Sportspiele Erleben, Vermitteln, Trainieren, K. Ferger, N. Gissel, and J. Schwier, Hrsg., Czwalina, Hamburg, 2002, S. 253–265]. In the following sections, some concepts and current approaches in the field of net-based data analysis are presented, and a case study demonstrates how it works in practice.

SOMSE: A semantic map based meta-search engine for the purpose of web information customization

To combat information overload, systems that are often referred to as information customization systems are needed. Such systems act on the user's behalf and can rely on existing information services like search engines that do the resource-intensive part of the work. These systems will be sufficiently lightweight to run on an average PC and serve as personal assistants. Since such an assistant has relatively modest resource requirements it can reside on an individual user's machine. If the assistant resides on the user's machine, there is no need to turn down intelligence. The system can have substantial local intelligence. In this paper, we propose an information customization system that combines meta-search and unsupervised learning. A meta-search engine simultaneously searches multiple search engines and returns a single list of results. The results retrieved by this engine can be highly relevant, since it is usually grabbing the first items from the relevancy-ranked list of hits returned by the individual search engines. The Kohonen Feature Map is then used to construct a self-organizing semantic map such that documents of similar contents are placed close to one another.

Visual Cluster Analysis of Trajectory Data with Interactive Kohonen Maps

Visual-interactive cluster analysis provides valuable tools for effectively analyzing large and complex data sets. Owing to desirable properties and an inherent predisposition for visualization, the Kohonen Feature Map (or Self-Organizing Map or SOM) algorithm is among the most popular and widely used visual clustering techniques. However, the unsupervised nature of the algorithm may be disadvantageous in certain applications. Depending on initialization and data characteristics, cluster maps (cluster layouts) may emerge that do not comply with user preferences, expectations or the application context. Considering SOM-based analysis of trajectory data, we propose a comprehensive visual-interactive monitoring and control framework extending the basic SOM algorithm. The framework implements the general Visual Analytics idea to effectively combine automatic data analysis with human expert supervision. It provides simple, yet effective facilities for visually monitoring and interactively controlling the trajectory clustering process at arbitrary levels of detail. The approach allows the user to leverage existing domain knowledge and user preferences, arriving at improved cluster maps. We apply the framework on several trajectory clustering problems, demonstrating its potential in combining both unsupervised (machine) and supervised (human expert) processing, in producing appropriate cluster results.

Dependency of recognition rate on number of words for text-independent speaker recognition using vector quantization

In this research, we discuss speaker recognition using the Kohonen feature map. The map is constructed for each speaker, and it is trained by using log-power and fourteenth-order mel-frequency cepstral coefficients (MFCC) and their temporal difference. The quantization distortion is computed between the input speech and a specific vector on the feature map of each speaker. We conduct speaker recognition experiment based on VQ distortion. Utterances of prefectural name in Japan are used as speech data. We examine particularly the dependency of recognition rate on number of words used for recognition. According to our experiments of speaker identification, this system correctly recognizes 98.9% by using a single word for 40 male speakers, while it attains 100% by using more than three words. Moreover, we confirmed superiority of using VQ over HMM under the same experimental conditions.

Effective palette indexing for image compression using self-organization of Kohonen feature map

The process of limited-color image compression usually involves color quantization followed by palette re-indexing. Palette re-indexing could improve the compression of color-indexed images, but it is still complicated and consumes extra time. Making use of the topology-preserving property of self-organizing Kohonen feature map, we can generate a fairly good color index table to achieve both high image quality and high compression, without re-indexing. Promising experiment results will be presented.

Medical image compression using topology-preserving neural networks

A novel method based on topology-preserving neural networks is used to implement vector quantization for medical image compression. The described method is an innovative image compression procedure, which differentiates itself from known systems in several ways. It can be applied to larger image blocks and represents better probability distribution estimation methods. A transformation-based operation is applied as part of the encoder on the block-decomposed image. The quantization process is performed by a “neural-gas” network which applied to vector quantization converges quickly to low distortion errors and reaches a distortion error lower than that resulting from Kohonen's feature map or the LBG algorithm. To study the efficiency of our algorithm, we blended mathematical phantom features into clinically proved cancer free mammograms. The influence of the neural compression method on the phantom features and the mammo-graphic image is not visually perceptible up to a high compression rate.

A neural network approach to movement pattern analysis

Movements are time-dependent processes and so can be modelled by time-series of coordinates: E.g., each articulation has geometric coordinates; the set of the coordinates of the relevant articulations build a high-dimensional configuration. These configurations – or “patterns” – give reason for analysing movements by means of neural networks: The Kohonen Feature Map (KFM) is a special type of neural network, which (after having been coined by training with appropriate pattern samples) is able to recognize single patterns as members of pattern clusters. This way, for example, the particular configurations of a given movement can be identified as belonging to respective configuration clusters, where the sequence of clusters to which the time-depending configurations belong, characterizes the process as a 2-dimensional trajectory. The advantages of this method are that • the high dimensionality of the original processes is reduced to two dimensional trajectories • the clusters are automatically determined by the network • all data for further analyses can automatically be transferred into a data base. Thus, the processes can either be visualized and analysed by an expert or again processed by further automatic analysing tools, as has been done with similarity matrices. The disadvantage is that a KFM-training needs a huge amount of information, which normally is not available from experiments. However, the Dynamically Controlled Network DyCoN (a special type of KFM) makes it possible to reduce the amount of original training data substantially – e.g., by adding stochastically generated ones. Currently, DyCoN is used in several projects in order to generally support analyses of processes in sport. It should be emphasized that the presented approach is not meant to improve the understanding or to develop models of human movement but to give a survey of the advantages and methodological aspects of net-based movement analysis.

Analysis of hippocampal atrophy in alcoholic patients by a Kohonen feature map.

We investigated the correlation of hippocampal volume with homocysteine, folate, vitamin B12 and B6 in alcoholic patients and healthy controls applying a Kohonen feature map (KFM) and conventional statistics. Representation of subjects on the KFM suggested an inverse correlation of hippocampal volume with blood levels of homocysteine and correlation with folate and vitamin B6. In conventional statistical analyses (t-test) reduced folate and increased homocysteine was found in alcoholics compared to healthy controls (p < 0.01). In female alcoholics vitamin B6 was reduced significantly (p = 0.03). Multiple linear regression analyses showed a significant correlation between average hippocampal volume and homocysteine (p < 0.001). KFM proved to be a sensitive tool for visualisation of statistical correlations in data sets even if no further statistical information is available.

Self-organizing feature maps for solving location–allocation problems with rectilinear distances

This study deals with solving uncapacitated location–allocation (LA) problems with rectilinear distances by using a method based on Kohonen self-organizing feature maps (SOFMs). By treating LA problems as clustering problems, this method has the advantage of extracting the structure of the input data by a self-organizing process based on adaptation rules. In this paper, a heuristic method is constructed by using SOFMs with a guided refining procedure, and its performance is compared with simulated annealing. The experimental results using the proposed guided refining procedure to reinforce the SOFM method show that the proposed method is excellent in terms of quality of solution and speed of computation. In addition, the experimental results suggest that SOFMs may provide an excellent approach when generating initial solutions for other heuristic or exact algorithms. This conjecture is made because most of the solutions yielded by SOFM are close to the optimal solution in all experiments. Scope and purpose Given the location of a set of customers with different demands, the LA problem is to select the locations of a number of supply centers to serve the customers and to decide the corresponding allocation of the customers to supply centers under a given optimization criterion. The LA problem may arise in fire stations, hospitals, police stations, telecommunication networks, and warehouses relative to production facilities and customers. In addition, the recent rapid growth of demand for supply chain management, which can decrease the total cost, while improving the quality of goods and services of various organizations, has drawn significant research attention. As an important part of supply chain management, the LA problem is related to the facility management service, logistics and distribution, order entry and customer service operations. Therefore, it should be indicated much more than before for organizations to improve their competitive advantage. In solving the LA problem, the costs of transportation between customers and supply centers are proportional to an appropriately determined distance, e.g. the Euclidean distance or rectilinear distance. The location of supply centers and the allocation of customers to supply centers must be considered simultaneously. The capacities of supply centers in LA problems can either be fixed or treated without any limitation if the capacity of each supply center can be adjusted according to the amount of allocated service demands of customers, which are referred to as capacitated and uncapacitated problems, respectively. In addition, the assumption of no interaction between supply centers makes LA problems different from p-hub problems. Currently, existing algorithms for solving the LA problems cannot meet the demands of rapidly changing industrial or market environments, since they are either time consuming or provide poor-quality solutions. Therefore, in this article, we propose a heuristic method using Kohonen's feature maps, which are artificial neural networks capable of mapping the distribution of input data to specified neurons, with a guided refining procedure to solve uncapacitated LA problems. In addition, the investigation of appropriateness of using Kohonen's feature maps (called SOFMs) as an initial solution-generating method for another heuristic or exact method is also provided.

Game analysis and control by means of continuously learning networks

The paper deals with the question, if and how the process of learning can be modelled, analysed and maybe improved by means of Neural Networks. The problem is that most of the developed types of network are designed for restricted technical purposes only. The similarity to neural brain structure is reduced to some basic formulae describing algorithms of static learning in the meaning of pattern recognition and activity selection.Using a more dynamic approach, however, it seems to be possible to model dynamic aspects of learning and decision processes as well. One step in this direction has been done with the Dynamically Controlled Network (DyCoN), which - in order to support continuous learning - has been developed on the basis of a conventional Kohonen Feature Map (KFM) and has been tested successfully in different areas of sport during the last couple of years.The approach presented here on the one hand shall demonstrate the performance that is induced by the ability of continuous learning. On the other hand, it is sketched in an outlook what could be future developments of this approach and what kind of basic work has to be done in order to better understand what is happening in a game.

Image processing with neural networks—a review

We review more than 200 applications of neural networks in image processing and discuss the present and possible future role of neural networks, especially feed-forward neural networks, Kohonen feature maps and Hopfield neural networks. The various applications are categorised into a novel two-dimensional taxonomy for image processing algorithms. One dimension specifies the type of task performed by the algorithm: preprocessing, data reduction/feature extraction, segmentation, object recognition, image understanding and optimisation. The other dimension captures the abstraction level of the input data processed by the algorithm: pixel-level, local feature-level, structure-level, object-level, object-set-level and scene characterisation. Each of the six types of tasks poses specific constraints to a neural-based approach. These specific conditions are discussed in detail. A synthesis is made of unresolved problems related to the application of pattern recognition techniques in image processing and specifically to the application of neural networks. Finally, we present an outlook into the future application of neural networks and relate them to novel developments.

The application of neural networks to vibrational diagnostics for multiple fault conditions

Vibration analysis has long been used for the detection and identification of machine fault conditions. The specific characteristics of the vibration spectrum that are associated with common fault conditions are quite well known, e.g. the BPOR spectral component reflecting bearing defects and the peak at the rotational frequency in the vibration spectrum indicating the degree of imbalance. The typical use of these features would be to determine when a machine should be taken out of operation in the presence of deteriorating fault conditions. Reliable diagnostics of deteriorating conditions may however be more problematic in the presence of simultaneous fault conditions. This paper demonstrates that the presence of a bearing defect makes it impossible to determine the degree of imbalance based on a single vibration feature, e.g. the peak at rotational frequency. In such a case, it is necessary to employ diagnostic techniques that are suited to the parallel processing of multiple features. Neural networks are the best known technique to approach such a problem. The paper demonstrates that a neural classifier using the X and Y components of both the peak at rotational frequency and the peak at BPOR frequency as input features can reliably diagnose the presence of bearing defect and can at the same time indicate the degree of imbalance. Several different neural techniques are evaluated for this purpose. It is first shown how Kohonen feature maps can be applied to do an exploratory analysis on the data, and to implement reliable classification of different bearing conditions. Different supervised neural classification techniques are then evaluated for their ability to reliably model the degree of imbalance, while also identifying the presence of defects.

Extracting Slipperiness Component from Weather and Traffic Data for Winter Maintenance Operations

Although traffic and weather information systems indicate existing air and road surface temperatures, many drivers want to know not the temperature but the degree of road slipperiness. The friction coefficient is the best index for snow and ice conditions, but it is difficult to calculate. Some weather condition data are closely correlated with the friction coefficient. In this study, the use of weather and traffic data to determine the degree of road slipperiness was examined quantitatively by analyzing field data obtained at an intersection on a trunk line. Two kinds of filters were effective in extracting the slipperiness component from the observed data—the Kohonen Feature Map (KFM) and principal component analysis (PCA). KFM kept the distribution of the observed data uniform by eliminating excessive data, and PCA eliminated multicollinearity among the observed variables. The first two principal components successfully represented the original observed data with an accumulated proportion of more than 0.85. A linear multiple regression model, in which the PCA score values were adopted as the explanatory variable, was also justified. The multiple correlation coefficient was satisfactory for many patterns, and the radiation-related data were effective in improving estimate precision.