Kohonen's Self-organizing Feature Map Research Articles

A modified self-organizing neural net for shape extraction

Some modifications on Kohonen's self-organizing feature map are discussed to make it suitable for finding skeletons of binary images. In Kohonen's feature map, the set of processors and their neighbourhoods are fixed and do not change in the learning process. This may pose problems when the set of input vectors represents a prominent shape. The reference vectors or weight vectors lying in zero-density areas are affected by input vectors from all the surrounding parts of the non-zero distribution [5]. Hence a shape extraction problem requires a dynamic change in the network topology. In the present paper, to overcome the limitations of Kohonen's feature maps, we propose a mechanism in which the set of processors and their neighbourhoods change adaptively during learning, to extract the shape of a binary object in the form of a skeleton.

Solving the AGV Problem via a Self-Organizing Neural Network

Automated Guided Vehicle (AGV)-based material handling systems (MHS) are used widely in Flexible Manufacturing Systems (FMS). The problem of AGV consists of the decisions and the operational control strategies of dispatching, routeing and scheduling of a set of AGVs under given system environments and operational objectives. One remaining challenge is to develop effective methods of AGV decisions for improved system productivity. This paper describes a prototype neural network approach for the AGV problem in an FMS environment. A group of neural network models are proposed to perform dispatching and routeing tasks for the AGV under conditions of single or multiple vehicles, and with or without time windows. The goal is to satisfy the transport requests in the shortest time and in a non-conflicting manner, subject to the global manufacturing objectives. Based on Kohonen's self-organizing feature maps, we have developed efficient algorithms for the AGVs decisions, and simulation results have been very encouraging.

A ROBUST ARTIFICIAL NEURAL NETWORK TECHNIQUE FOR DYNAMIC STABILITY ASSESSMENT

ABSTRACT This paper presents a robust artificial neural network technique to assess the on-line dynamic stability of power system. Efforts have been made to minimize the error in dynamic stability assessment by optimising the size and introducing more sensitive information in training vector. The problem has been attempted with Kohonen's self-organising feature map (SOFM) to classify the states of power system. The main reason which inspired the authors to apply Kohonen's SOFM technique is to avoid local minima that saturate the learning process in back-propagation algorithm. Conventional QR algrithm in conjunction with S-matrix method is used to allocate dynamic stability indices to output neurons.

Multiple shell structured hypercube feature maps for vector quantization of images

Abstract A new neural network architecture is proposed for spatial domain image vector quantization (VQ). The proposed model has a multiple shell structure consisting of binary hypercube feature maps of various dimensions, which are extended forms of Kohonen's self-organizing feature maps (SOFMs). It is trained so that each shell can contain similar-feature vectors. A partial search scheme using the neighborhood relationship of hypercube feature maps can reduce the computational complexity drastically with marginal coding efficiency degradation. This feature is especially proper for vector quantization of a large block or high dimension. The proposed scheme can also provide edge preserving VQ by increasing the number of shells, because shells far from the origin are trained to contain edge block features.

Well-Test Model Identification With Self-Organizing Feature Map

Summary Well-test data have been used traditionally for determining a variety of reservoir parameters, such as average permeability, storage capacity, reservoir damage, presence of faults and fractures, and reservoir mechanism. A number of techniques, both conventional methods, such as type-curve matching and numerical simulation, and artificial intelligence (AI) methods, have been used for identifying well-test models. These methods are laborious and time-consuming and at times give incorrect results. Artificial neural networks (ANN's) are recent developments in computer vision and image analysis. These networks are specialized computer software that generate a strategy to produce nonlinear mapping functions for complex problems. ANN's are commonly used as a tool for recognizing an object or predicting an event given an associated pattern. Only a limited number of applications of ANN for analyzing well-test data have been reported. These applications are mostly model-specific (developed for specific reservoir models) and, hence, are not general enough. This paper presents a new method based on ANN's that uses Ko-honen's self-organizing feature (SOF) mapping technique to identify well-test interpretation models. By grouping well-test data into distinct categories, the SOF algorithm produces a general mapping function. This method can help analyze well-test data from a large variety of reservoirs (including reservoirs with faults, fractures, boundaries, etc.) more efficiently and inexpensively than was feasible previously.

Discrimination between local microearthquakes and quarry blasts by multi-layer perceptrons and Kohonen maps

Abstract The results of the application of artificial neural nets (ANNs) to discriminating microearthquakes from quarry and mining blasts in the West Bohemia earthquake swarm region are presented and discussed. Input vectors consisting of seven spectral and seven amplitude parameters, automatically extracted from local three-component digital broadband (0.6 to 60 Hz) velocigrams, have been employed for training of different ANN configurations. Multi-layer perceptrons (MLP) trained in supervised mode by different subsets of a representative set of 312 events have been used as discriminators, and unsupervised Kohonen self-organizing feature maps (SOFM) have been used as complementary reliability estimators. The reason for comparative application of both techniques was to increase the reliability of the discrimination: complementary information that a pattern has been recognized as a member of a conflict cluster allows detecting problematic patterns that an MLP may not be able to classify correctly. The optimal MLP, trained by one randomly selected half of the complete set of 312 input vectors and tested by the other half-set, and vice versa, correctly classified, on average, 99% of all events. The optimal SOFM correctly classified as problematic patterns all events misinterpreted by the MLP, and about 20% of all events were classified by them as ambiguous cases. The obtained results evidence that a relatively small number of spectral and amplitude parameters of observed ground velocity may suffice for a reliable discrimination between local microearthquakes and quarry blasts by means of small neural nets. The MLP/SOFM combination discussed in this article has attained a discrimination reliability that allows it to be employed routinely in observatory practice.

A neural network model for the free-ranging AGV route-planning problem

This paper describes the development of a prototype neural network model for the free-ranging AGV route-planning problem. The vehicle planner operates in quasi-real time. A small planning horizon is set and all transport requests existing at the beginning of a planning horizon are examined. A neural network model is proposed to perform dispatching and routing tasks for the AGVs. Its goal is to satisfy the transport requests in the shortest time and in a non-conflicting manner, subject to the global manufacturing objective of maximizing throughputs. Based on Kohonen's self-organizing feature maps, we develop three efficient planning algorithms for the single and multiple AGV problems. The simulation results indicate that the proposed neural network approach gives very efficient solutions.

Controlling the Magnification Factor of Self-Organizing Feature Maps

The magnification exponents μ occurring in adaptive map formation algorithms like Kohonen's self-organizing feature map deviate for the information theoretically optimal value μ = 1 as well as from the values that optimize, e.g., the mean square distortion error (μ = 1/3 for one-dimensional maps). At the same time, models for categorical perception such as the "perceptual magnet" effect, which are based on topographic maps, require negative magnification exponents μ < 0. We present an extension of the self-organizing feature map algorithm, which utilizes adaptive local learning step sizes to actually control the magnification properties of the map. By change of a single parameter, maps with optimal information transfer, with various minimal reconstruction errors, or with an inverted magnification can be generated. Analytic results on this new algorithm are complemented by numerical simulations.

Comparison of several approaches for the segmentation of texture images

In this paper, several approaches including K-Means, Fuzzy K-Means (FKM), Fuzzy Adaptive Resonance Theory (ART2) and Fuzzy Kohonen Self-Organizing Feature Mapping (SOFM) are adapted to segment the texture image. In our tests, five features, energy, entropy, correlation, homogeneity, and inertia, are used in texture analysis. The K-Means algorithm has the following disadvantages: (i) slow real-time ability, (ii) unstability. The FKM algorithm has improved the performance of the unstability by means of the introduction of fuzzy distribution functions. The Fuzzy ART2 has advantages, such as unsupervised training, low computation, and great degree of fault tolerance (stability/plasticity). Fuzzy operator and mapping functions are added into the network to improve the generality. The Fuzzy SOFM integrates the FKM algorithm into fuzzy membership value as learning rate and updating strategies of the Kohonen network. This yields automatic adjustment of both the learning rate distribution and update neighborhood, and has an optimization problem related to FKM. Therefore, the Fuzzy SOFM is independent of the sequence of feed of input patterns whereas final weight vectors by the Kohonen method depend on the sequence. The Fuzzy SOFM is “self-organizing” since the “size” of the update neighborhood and learning rate are automatically adjusted during learning. Clustering errors are reduced by Fuzzy SOFM as well as better convergence. The numerical results show that Fuzzy ART2 and Fuzzy SOFM are better than the K-Means algorithms. The images segmented by the algorithms are given to prove their performances.

Internet Categorization and Search: A Self-Organizing Approach

The problems of information overload and vocabulary differences have become more pressing with the emergence of increasingly popular Internet services. The main information retrieval mechanisms provided by the prevailing Internet WWW software are based on either keyword search (e.g., the Lycos server at CMU, the Yahoo server at Stanford) or hypertext browsing (e.g., Mosaic and Netscape). This research aims to provide an alternative concept-based categorization and search capability for WWW servers based on selected machine learning algorithms. Our proposed approach, which is grounded on automatic textual analysis of Internet documents (homepages), attempts to address the Internet search problem by firstcategorizingthe content of Internet documents. We report results of our recent testing of a multilayered neural network clustering algorithm employing the Kohonen self-organizing feature map to categorize (classify) Internet homepages according to their content. The category hierarchies created could serve to partition the vast Internet services into subject-specific categories and databases and improve Internet keyword searching and/or browsing.

Feature extraction and pattern classification of remote sensing data by a modular neural system

A modular neural network architecture has been used for the classification of remote sensed data in two experiments carried out to study two different but rather usual situations in real remote sensing applications. Such situations concern the availability of high-dimensional data in the first setting and an imperfect data set with a limited number of features in the second. The learning task of the supervised multilayer perceptron classifier has been made more efficient by preprocessing the input with unsupervised neural modules for feature discovery. The linear propagation network is introduced in the first experiment to evaluate the effectiveness of the neural data compression stage before classification, whereas in the second experiment data clustering before labeling is evaluated by the Kohonen self-organizing feature map network. The results of the two experiments confirm that modular learning performs better than nonmodular learning with respect to both learning quality and speed. © 1996 Society of Photo-Optical Instrumentation Engineers. Subject terms: neural network; remote sensing; classification; supervised and un- supervised learning.

<b>Kohonen</b>自己組織化特徴マップを用いた巡回セールスマン問題解法の改良

<b>Kohonen</b>自己組織化特徴マップを用いた巡回セールスマン問題解法の改良

Self-organizing feature map with a momentum term

The objectives of this paper are to derive a momentum term in the Kohonen's self-organizing feature map algorithm theoretically and to show the effectiveness of the term by computer simulations. We will derive the self-organizing feature map algorithm having the momentum term through the following assumptions: (1) The cost function is E n = Σ μ n α n − μ E μ , where E μ is the modified Lyapunov function originally proposed by Ritter and Schulten at the μth learning time and α is the momentum coefficient. (2) The latest weights are assumed in calculating the cost function E n . According to our simulations, it has shown that the momentum term in the self-organizing feature map can considerably contribute to the acceleration of the convergence.

Topographic map formation by maximizing unconditional entropy: a plausible strategy for "online" unsupervised competitive learning and nonparametric density estimation

An unsupervised competitive learning rule, called the vectorial boundary adaptation rule (VBAR), is introduced for topographic map formation. Since VBAR is aimed at producing an equiprobable quantization of the input space, it yields a nonparametric model of the input probability density function. Furthermore, since equiprobable quantization is equivalent to unconditional entropy maximization, we argue that this is a plausible strategy for maximizing mutual information (Shannon information rate) in the case of "online" learning. We use mutual information as a tool for comparing the performance of our rule with Kohonen's self-organizing (feature) map algorithm.

Neural networks and information extraction in astronomical information retrieval

We describe results obtained in prototyping a Kohonen self-organising feature map as a user interface to a large document collection. We then discuss the provision of various tools which will be used in quality control of databases and catalogs produced from, and cross-referenced with, the online astronomical literature; and which will be used in future information seeking and processing intelligent agent technologies.

Classification of binary document images into textual or nontextual data blocks using neural network models

This paper describes a new method for the classification of binary document images as textual or nontextual data blocks using neural network models. Binary document images are first segmented into blocks by the constrained run-length algorithm (CRLA). The component-labeling procedure is used to label the resulting blocks. The features for each block, calculated from the coordinates of its extremities, are then fed into the input layer of a neural network for classification. Four neural networks were considered, and they include back propagation (BP), radial basis functions (RBF), probabilistic neural network (PNN), and Kohonen's self-organizing feature maps (SOFMs). The performance and behavior of these neural network models are analyzed and compared in terms of training times, memory requirements, and classification accuracy. The experiments carried out on a variety of medical journals show the feasibility of using the neural network approach for textual block classification and indicate that in terms of both accuracy and training time RBF should be preferred.

Neural network models in EMG diagnosis

In previous years, several computer-aided quantitative motor unit action potential (MUAP) techniques were reported. It is now possible to add to these techniques the capability of automated medical diagnosis so that all data can be processed in an integrated environment. In this study, the parametric pattern recognition (PPR) algorithm that facilitates automatic MUAP feature extraction and Artificial Neural Network (ANN) models are combined for providing an integrated system for the diagnosis of neuromuscular disorders. Two paradigms of learning for training ANN models were investigated, supervised, and unsupervised. For supervised learning, the back-propagation algorithm and for unsupervised learning, the Kohonen's self-organizing feature maps algorithm were used. The diagnostic yield for models trained with both procedures was similar and on the order of 80%. However, back propagation models required considerably more computational effort compared to the Kohonen's self-organizing feature map models. Poorer diagnostic performance was obtained when the K-means nearest neighbor clustering algorithm was applied on the same set of data.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Clustering properties of fuzzy Kohonen's self-organizing feature maps

A new clustering algorithm called fuzzy self-organizing feature maps is introduced. It can process not only the exact digital inputs, but also the inexact or fuzzy non-digital inputs, such as natural language inputs. Simulation results show that the new algorithm is superior to original Kohonen's algorithm in clustering performance and learning rate.

A hybrid learning neural network architecture with locally activated hidden layer for fast and accurate mapping

A hybrid learning neural network architecture based on global clustering and local learning is developed that not only speeds up learning but also enhances mapping accuracy for both real-valued and logic functions. This fast learning while satisfying a prescribed accuracy is an essential characteristic for real-time on-line learning systems. Either Kohonen's self-organizing feature map or the leader clustering algorithm is used to partition the input data space for local learning. The input data selects a subset of hidden nodes (either sigmoidal or Gaussian) that contribute to the output calculation. Example results demonstrate the proposed architecture's superior convergence properties over the original backpropagation network or its improvement techniques.

Interpreting the Kohonen self-organizing feature map using contiguity-constrained clustering

An interpretation phase is proposed, to complement usage of the Kohonen self-organizing feature map (SOFM) method. This segments the SOFM output, using an agglomerative contiguity-constrained clustering method. We discuss why such a clustering method, which respects contiguity information, should be used. The contiguity-constrained clustering method implemented uses array operations as far as possible, and is thus quite efficient. An application to an astronomical catalog of about a quarter million objects is presented.