Self Organized Map

During the last several years, lidar has become a widely used technique for data collection from the earth surface and vegetation canopy being the large volume of high density lidar data the main drawback for its interpretation and analysis. In addition, parcelbased segmentation of high-resolution remotely sensed data can provide convenient and useful spatial and structural information. In this paper, a methodology for semi-automatic updating of forest land use/land cover geo-spatial databases, using high spatial resolution imagery and lidar data, is presented. High spatial resolution multispectral imagery and low density lidar data (0.5 points/m 2 ) has been employed. Cartographic limits from a cadastre geospatial database have been used in order to segment the territory and create analysis objects. The objects are characterized using a set of descriptive features: spectral, structural, shape and texture features computed from the multispectral image. These features are combined with 3D features derived from lidar data: density profiles based indices and statistics from point cloud, intensity values and normalized digital surface models. The lidar descriptive features proposed provide a more intuitive interpretation of the vegetation canopy structure than the raw data. The classification is performed using the decision trees technique combined with the boosting multi-classifier. Classification assessment is done by using ground truth data.

This paper proposes that abstract concepts are represented as contextually derived structures. According to this abstract structure theory, abstract concepts are related to mostly temporal and spatial structures that underlie and can be extracted from concrete situations. Linguistic context elements of abstract concepts, such as verbs and prepositions, express these structures, and can thus aid in the acquisition of such concepts. The paper presents results from a corpus study that supports the hypothesis, and discusses implications. We propose that abstract concepts, such as faith or notion, are represented as abstract structures, which represent particular contexts in which they occur. Abstract concepts are not directly perceivable, but are often used in verbal descriptions of situations, or in utterances related to a situation. Such utterances have to unambiguously point out the entity that is referred to by the abstract noun. We argue that the relevant abstract structures can be inferred from their linguistic context, in particular, from verbs and prepositions used with the abstract nouns. Similar concepts occur in similar linguistic contexts. Miller and Charles (1991) report that contexts of similar concepts are more often classified as belonging to the same concept. For example, the sentences the patient rang for the ____ and the ____ gave the patient an injection both suggest that the word doctor or nurse would complete the sentence well. We found that a neural network could be trained to correctly select one out of seven abstract concepts based on linguistic context information in 72% of test cases (Wiemer-Hastings, 1998). An approach to learning verbs from context was described by Wiemer-Hastings, Graesser, and Wiemer-Hastings (1998; see also Hastings, 1994). This work shows a close link between context of use and concepts.

Visual Analytics seeks to combine automatic data analysis with visualization and human-computer interaction facilities to solve analysis problems in applications characterized by occurrence of large amounts of complex data. The financial data analysis domain is a promising field for research and application of Visual Analytics technology, as it prototypically involves the analysis of large data volumes in solving complex analysis tasks. We introduce a Visual Analytics system for supporting the analysis of large amounts of financial time-varying indicator data. A system, driven by the idea of extending standard technical chart analysis from one to two-dimensional indicator space, is developed. The system relies on an unsupervised clustering algorithm combined with an appropriately designed movement data visualization technique. Several analytical views on the full market and specific assets are offered for the user to navigate, to explore, and to analyze. The system includes automatic screening of the potentially large visualization space, preselecting possibly interesting candidate data views for presentation to the user. The system is applied to a large data set of time varying 2-D stock market data, demonstrating its effectiveness for visual analysis of financial data. We expect the proposed techniques to be beneficial in other application areas as well.

Low-dimensional (2-or 3-dimensional) visual representations of large, highdimensional datasets with complicated cluster structures play a fundamental role in the discovery and identification of such structures. Visualization exploits the unmatched pattern recognition capability of humans for accurate and detailed cluster extraction, which is not possible with current automated methods because the latter still lack the power of the exceptional human reasoning. For explanatory and interactive visualization, a powerful tool is the use of self-organizing maps (SOMs). In general, by producing a spatially ordered set of quantization prototypes of large, higher-dimensional data, SOMs enable the visualization of various similarity information (such as prototype distances, distribution, topology) on a rigid lattice, without reducing the feature dimensionality. Information discovery further depends on the expressive power of the similarity measure and its visual representation. In this study, we compare the capabilities of our recent SOM visualization scheme, CONNvis, with prominent dimensionality reduction methods and show its superiority for visual assessment of intricate cluster structures.

Resumen Las redes neuronales artificiales (RNA) son aplicadas en diversos ámbitos de la actividad humana. Una de sus aplicaciones es como herramienta de análisis de información, específicamente dentro de la Bibliometría. En este trabajo se hace una ...

The purpose of this paper is to introduce Artificial Intelligence in the field of data-security and to propose an easy to implement Neural Networks based method for user authentication. The problem has been faced exploiting an RBF-like Neural Net to recognize the typing style of users asking for connection. The introduction of Neural Nets allows to extract rules directly from row data (users typing times) without any assumption on these.

Growing self-organizing map (GSOM) has been introduced as an improvement to the self-organizing map (SOM) algorithm in clustering and knowledge discovery. Unlike the traditional SOM, GSOM has a dynamic structure which allows nodes to grow reflecting the knowledge discovered from the input data as learning progresses. The spread factor parameter (SF) in GSOM can be utilized to control the spread of the map, thus giving an analyst a flexibility to examine the clusters at different granularities. Although GSOM has been applied in various areas and has been proven effective in knowledge discovery tasks, no comprehensive study has been done on the effect of the spread factor parameter value to the cluster formation and separation. Therefore, the aim of this paper is to investigate the effect of the spread factor value towards cluster separation in the GSOM. We used simple k-means algorithm as a method to identify clusters in the GSOM. By using Davies-Bouldin index, clusters formed by different values of spread factor are obtained and the resulting clusters are analyzed. In this work, we show that clusters can be more separated when the spread factor value is increased. Hierarchical clusters can then be constructed by mapping the GSOM clusters at different spread factor values.

This paper introduces a novel approach to Self-Organizing Maps which is capable of processing graphs such that the context of vertices and sub-graphs are considered in the mapping process. The result is that any vertex in a graph is mapped onto an n-dimensional map depending on its label and the graph structure as a whole. Experimental results demonstrate that the proposed approach achieves the desired outcomes.

For the latest ten years, many authors have focused their investigations in wireless sensor networks. Different researching issues have been extensively developed: power consumption, MAC protocols, selforganizing network algorithms, data-aggregation schemes, routing protocols, QoS management, etc. Due to the constraints on data processing and power consumption, the use of artificial intelligence has been historically discarded. However, in some special scenarios the features of neural networks are appropriate to develop complex tasks such as path discovery. In this paper, we explore the performance of two very well known routing paradigms, directed diffusion and Energy-Aware Routing, and our routing algorithm, named SIR, which has the novelty of being based on the introduction of neural networks in every sensor node. Extensive simulations over our wireless sensor network simulator, OLIMPO, have been carried out to study the efficiency of the introduction of neural networks. A comparison of the results obtained with every routing protocol is analyzed. This paper attempts to encourage the use of artificial intelligence techniques in wireless sensor nodes.

A complete framework for exciting and detecting thermally-induced, stabilized sine-Gordon breathers in ac-driven long Josephson junctions is developed. The formation of long-time stable breathers locked to the ac source occurs for a sufficiently high temperature. The latter emerges as a powerful control parameter, allowing for the remarkably stable localized modes to appear. Nonmonotonic behaviors of both the breather generation probability and the energy spatial correlations versus the thermal noise strength are found. The junction's resistive switching characteristics provides a clear experimental signature of the breather.

Horizontal and vertical integration of engineering education is achieved through an early-design project where students get acquainted with Total Quality Management (TQM) principles and design processes from year-one of their University education. The project is embedded in the undergraduate chemical engineering curriculum as an activity that involves horizontally several first-year subjects and vertically a fourth-year Project Management Practice course and a related Project Management subject. An assessment of the integrated design project indicates that effective teaching and learning spreads through the curriculum, with fourth-year students acting as project managers and experiencing engineering practice. These management and leadership training processes include a shared responsibility in the organization and in the development of the project, which are key factors for the success of the integrated activity. They are also a first step towards the ETSEQ goal of becoming a sustainable student-centered educational system.

Self-organized maps (SOM) have been applied to analyze the similarities of chemical compounds and to select from a given pool of descriptors the smallest and more relevant subset needed to build robust QSAR models based on fuzzy ARTMAP. First, the category maps for each molecular descriptor and for the target activity variable were created with SOM and then classified on the basis of topology and nonlinear distribution. The best subset of descriptors was obtained by choosing from each cluster the index with the highest correlation with the target variable and then in order of decreasing correlation. This process was terminated when a dissimilarity measure increased, indicating that the inclusion of more molecular indices would not add supplementary information. The optimal subset of descriptors was used as input to a fuzzy ARTMAP architecture modified to effect predictive capabilities. The performance of the integrated SOM-fuzzy ARTMAP approach was evaluated with the prediction of the acute toxicity LC 50 of a homogeneous set of 69 benzene derivatives in the fathead minnow and the oral rat toxicity LD 50 of a heterogeneous set of 155 organic compounds. The proposed methodology minimized the problem of misclassification of similar compounds and significantly enhanced the predictive capabilities of a properly trained fuzzy ARTMAP network.

A new approach is presented for the development of quantitative structure–property relations (QSPR) based on the extraction of relevant molecular features with self‐organizing maps and the use of a modified fuzzy‐ARTMAP classifier for variable prediction. The present methodology is demonstrated for the development of a QSPR for the aqueous‐phase infinite dilution activity coefficient γ∞, based on a data set of 325 diverse organic compounds. The QSPR was developed using a set of 11 molecular descriptors (four connectivities vχ1–4, Coulomb self‐similarity measure, electron–nuclear attraction, dipole moment, sum of atomic numbers, number of filled levels, average polarizability, and nuclear–nuclear repulsion). The final set of molecular descriptors was selected from an initial pool of 23 topological and quantum chemical descriptors, including six molecular quantum similarity measures, by means of a topological analysis of self‐organization of the data set. Additional interpolated infor...

Organic solute permeation, sorption, and rejection by reverse osmosis membranes, from aqueous solutions, were studied experimentally and via artificial neural networks (ANN)-based quantitative structure-property relations (QSPR), for a set of fifty organic compounds for polyamide and cellulose acetate membranes. Membrane solute sorption and passage for dead-end filtration model experiments were quantified based on radioactivity measurements for radiolabeled compounds in the feed, permeate and the membrane, while solute rejection was determined from a mass balance on the permeated solution volume. Artificial neural networks-based quantitative structure-property relations models were developed for the organic passage (P), sorbed (M) and rejected (R) fractions using the most relevant set of molecular descriptors selected from a pool of 45 molecular descriptors by means of a correlation-based feature selection method and self-organizing maps (SOM). The analysis included pre-screening with principal components analysis and SOM of the chemical domain for the study chemicals as defined by chemical descriptors to identify the applicability domain and chemical similarities. The QSPR models predicted the P and M mass fractions within the range of the standard deviations of measurements for the experimental data set of fifty compounds. Mass balance closure (requiring that M, P and R sum to unity) was satisfactory for the experimental data set of fifty compounds and for an external set of 144 test chemicals, which were not included in the model development. Somewhat higher prediction errors were encountered for a few chemicals that were not well represented within the present chemical domain. The quality of the QSPR/NN models developed suggests that there is merit in extending both the present compound database and the present approach to develop a comprehensive tool for assessing organic solute behavior in RO water treatment processes.

A predictive Fuzzy ARTMAP neural system and two hybrid networks, each combining a dynamic unsupervised classifier with a different kind of supervised mechanism, were applied to develop virtual sensor systems capable of inferring the properties of manufactured products from real process variables. A new method to construct dynamically the unsupervised layer was developed. A sensitivity analysis was carried out by means of self-organizing maps to select the most relevant process features and to reduce the number of input variables into the model. The prediction of the melt index (MI) or quality of six different LDPE grades produced in a tubular reactor was taken as a case study. The MI inferred from the most relevant process variables measured at the beginning of the process cycle deviated 5% from on-line MI values for single grade neural sensors and 7% for composite neural models valid for all grades simultaneously.

Abstract: In this paper we describe a neural network for the nonlinear adaptive prediction of non-stationary signals and demonstrate its application to a speech signal. The network is a multi-layer perceptron trained with the backpropagation algorithm. The detail of this ...

Segmentation is a fundamental step in image description or classiÿcation. In recent years, several computational models have been used to implement segmentation methods but without establishing a single analytic solution. However, the intrinsic properties of neural networks make them an interesting approach, despite some measure of ine ciency. This paper presents a clustering approach for image segmentation based on a modiÿed fuzzy approach for image segmentation (ART) model. The goal of the proposed approach is to ÿnd a simple model able to instance a prototype for each cluster avoiding complex post-processing phases. Results and comparisons with other similar models presented in the literature (like self-organizing maps and original fuzzy ART) are also discussed. Qualitative and quantitative evaluations conÿrm the validity of the approach proposed.

In the last years, the interest for advanced video-based surveillance applications is more and more growing. This is especially true in the field of railway urban transport where video-based surveillance can be exploited to face many relevant security aspects (e.g. vandal acts, overcrowding situations, abandoned object detection, etc.). This paper 1 aims at investigating an open problem in the implementation of videobased surveillance systems for transport applications, i.e.: the implementation of reliable image understanding modules in order to recognize dangerous situations with reduced false alarm and misdetection rates. In this work, we considered the use of a neural network-based classifier for detecting vandal behaviors in metro stations. The achieved results show that the classifier choice mentioned above allows one to achieve very good performances also in presence of high scene complexity.

The weather during grape production affects wine quality. Changes in the weather in the Chablis region of France and in the quality of Chablis wines (vintage scores) from 1963 to 2018 were analysed. Chablis wine quality improved over this period, with no Poor vintages after 1991. Summer temperature and sunshine duration both increased progressively between 1963 to 2018 with fewer frost days but no linear change detected in precipitation. Chablis vintage score was modelled as a function of mean temperature from April to September (curvilinear relation, maximum score at 16–17 °C), mean minimum temperature in September (an index of cool nights; negative relation), and total rainfall from June to September (negative relation). This simple three-factor model distinguished between Poor and higher-quality Chablis vintages well, but less so between Good and Excellent vintages. Application of the model to different climate change scenarios (assuming current viticultural and oenological pract...

The VICOM (Virtual Immersive COMmunications) project is a three-year project funded by the Italian Ministry of Instruction University and Research aiming at investigating innovative communication paradigms. The project represents a wide coordinated effort focused on integration of immersive and wireless technologies in view of the fourth generation of mobile communications. The main goal of the project consists of the design of a wideband system architecture for immersive services and of its validation through two distributed large test-beds. Starting from VICOM ongoing experiences some future challenges and objectives for the future situated and autonomic communications technologies are envisaged in the paper.

The objective of this research is to show the capacity of Self-Organizing Maps to classify customer and their response potential from electrical demand databases with the help of Non-Parametric Estimation and Physically Load Based modelling as support tools. The searching of customer suitability is focussed to real time products, whose interest is growing in developed countries. In this way customer demand and response have been tested and compared with energy price curves extracting patterns from these curves. Results show the capability of this approach to improve data management and select coherent policies to accomplish cleared demand offers amongst different prices scenarios in an easy way.

Information and Communication Technology (ICT) is one of the most inportant aspect in a country. Good progress in ICT will be a valuable factor to compete with other countries. The progress in DKI Jakarta province is considered as a measurement or representation, that the ICT in Indonesia has been developed well or not. This research will use data mining method with clustering, using Self Organizing Map (SOM) algorithm. The algorithm is used for clustering the villages in DKI Jakarta Province, based on the availability of telecommunication and internet facility. The clustering or grouping result, less or more, will describe the progress rate in the province. Last, the ICT progress in the country hoped will be represented in village clustering data of DKI Jakarta.

In this paper, a hierarchial neural network architecture for forecasting time series is presented. 7he architecture is ampmd of two hierarchial CGVclr using a maximum likelihood competitive learning algorithm. lhefirsr Lcvd of the systan hpr thtcc cqwrts eiach using ba&"gation and a gating network to partition the input space in order to map the input wctors to the output wctors. llhc second lewl of the hiearchial network has an expert using Fuuy ARTfot producing the comct trend d n g j b m thefirst level. ?%e experiments show that the resulting network is qmbk offmcmting the changes in the input and identifling the t r e h correctly,

In this article, the quantity of grapes sold in one fruit shop of an interlocking fruit supermarket is forecasted by the method of support vector machine (SVM) based on deficient data. Since SVMs have a lot advantages such as great generalization performance and guarantying global minimum for given training data, it is believed that support vector regression will perform well for forecasting sales of grapes. In order to improve forecasting precision (FP), this article quantifies the factors affecting the sales forecast of grapes such as weather and weekend or weekday, results are suitable for real situations. In this article, we apply ε -SVR and LS-SVR to forecast sales of three varieties of grapes. Moreover, the artificial neural network (ANN) and decision tree (DT) are used as contrast and numerical experiments show that forecasting systems with SVMs is better than ANN and DT to forecast the daily sales of grapes overall.

The effectiveness of a computer-based method for localization of arrhythmia exit sites was studied. The proposed algorithm works on any set of 3 or more ECG leads. The QRS complex integral of an ectopic beat is reduced to principal components treated as coordinates of the exit site in ECG space and then projected to real space by a linear transformation. The accuracy of the method was tested on 5 patient-tailored models of human heart and torso. For each model ~500 simulations were run, each for different stimulus location. All locations were then estimated from simulated surface ECGs and method accuracy was investigated. The algorithm performed better for the left ventricle than for the right ventricle. The group mean absolute and relative (to a neighboring stimulation site) localization errors in millimeters were: 11.5 (SD=2.5), 2.6 (SD=0.5) for the 252-lead ECG; 12.2 (SD=2.7), 2.7 (SD=0.5) for the and 11.7 (SD=2.4), 2.7 (SD=0.5) for the Frank VCG. This study suggest that the proposed method can predict exit sites with a precision in the order of a centimeter. Low values of error for neighbouring activation sites suggest opportunity for algorithm improvement. The use of vectorcardiographic leads is enough to obtain a precision comparable to a 252-lead ECG.

Recommender systems are tools to help users find items that they deem of interest to them. They can be seen as an application of data mining process. In this paper, a new recommender system based on multi-features is introduced. Demographic and psychographic features are used to asses similarities between users. The model is built on a collaborative filtering method and addresses three problems: sparsity, scalability and cold-star. The sparsity problem is tackled by integrating users-documents relevant information within meta-clusters. The scalability and the cold-start problems are considered by using a suitable probability model calculated on meta-cluster information. Moreover, a weight similarity measure is introduced in order to take into account dynamic human being preferences behaviour. A prediction score for generating recommendations is proposed based on the target user previous behaviour and his/her neighbourhood preferences on the target document.

Spatial fields of outgoing long wave radiation (OLR) spectrum variance of the 1979-2016 austral summer months in southern Brazil are analyzed on different timescales: synoptic, sub-monthly, and intra-seasonal. Variability fields differ both in intensity and location and highlight dominant convection cycles in the study area. The results show that the amplitude of sub-monthly variability is greater than that of the other scales in the southeastern region of Brazil, while the synoptic scale prevails in the southern region. The above-mentioned scales show greater amplitudes over the western Pacific Ocean where the Madden-Julian Oscillation plays an important role, along the South Pacific Convergence Zone, and over the storm track areas over the South Pacific Ocean. The influence of spectral OLR scale interaction is also analyzed, associated to the occurrence of two intense rainfall events over the southeastern Brazil in the austral summers of 2011 and 2014 when the South Atlantic Conve...

In previous work, cynaropicrin and grosheimin derivatives were submitted to a panel test for sensory evaluation. Bitterness variations seemed to be related to changes in molecular polarity. Reported incongruences have confused the understanding of the molecular mechanisms of bitterness variations. In this work, a classic QSAR analysis was applied to a training set of 15 sesquiterpene lactones and the structure-bitterness relationship investigated. The dipole moment, polarizability and hydrophobicity were studied and made it possible to conclude that bitterness of sesquiterpene lactones was not directly related to molecular polarity. The calculated molecular descriptors were used, in combination with classic QSAR, to define the relevant parameters responsible for the biological properties. The variable selection showed a satisfactory quantitative structure-activity relationship (QSAR) and a model was established.

This chapter examines abstract concepts of innovators' competences and innovation culture. For people to be innovative, both concepts need to be considered. Ontologies provide a way to specify these abstract concepts into such a format that practical applications can be applied in organizations. Self-evaluation of innovation competence and innovation culture in organizations can be conducted by utilizing a fuzzy logic application platform called Evolute. The approach described in this chapter has management implications. The abstract concepts of innovation culture and innovation competence become manageable, which suggests that organizations should be able to get better innovation results.

The evolutionary history of Neotropical crocodiles has remained elusive. They inhabit a broad geographic range with populations spanning from coastal, inland, and insular locations. Using a selection of natural insular, coastal, and one inland population of C. acutus, coastal C. moreletii, and the single surviving population of C. rhombifer, we discovered a remarkable genetic diversity for the group. Moreover, geometric morphometric results of skull shapes shows that these crocodylus species span a morphological cline. We recovered a high genetic differentiation between C. moreletii, C. rhombifer, and five clusters of C. acutus. The genetic and geographic differences among the C. acutus clusters were used to suggest these may be a species complex. Several ecological, morphological and genetics traits are identified in the well-studied populations from Banco Chinchorro and Cozumel islands off the Mexican Yucatan Peninsula to support discrete species designations for these populations. This work suggests the presence of rapid, recent evolution of several cryptic Crocodylus species throughout the Neotropics.

Pan-tilt-verge (PTV) vision system is one of the most widely used in active vision. The main advantage of using such system is its 4 DOF which allows tracking of moving objects efficiently. Besides a physical design of the head, an overall tracking performance of the system depends on its controller. This paper presents a development of controlling PTV head to achieve one of human-like eye movement behaviors, i.e. saccade. The PTV head is driven directly from a controller using visual feedback. The dynamic Jacobian estimation is obtained by using a self-organizing map network with unsupervised learning scheme. The estimated Jacobian is used in PTV head controller and results are desirable for both performance and speed of learning. Moreover, the system can eventually perform tracking without a priori knowledge of the head structure, e.g. mathematical model of the head and hardware calibration. Thus the system can conveniently be implemented.

Pan-tilt-verge (PTV) vision system is one of the most widely used in active vision. The main advantage of using such system is its 4 DOF which allows tracking of moving objects efficiently. Besides a physical design of the head, an overall tracking performance of the system depends on its controller. This paper presents a development of controlling PTV head to achieve one of human-like eye movement behaviors, i.e. saccade. The PTV head is driven directly from a controller using visual feedback. The dynamic Jacobian estimation is obtained by using a self-organizing map network with unsupervised learning scheme. The estimated Jacobian is used in PTV head controller and results are desirable for both performance and speed of learning. Moreover, the system can eventually perform tracking without a priori knowledge of the head structure, e.g. mathematical model of the head and hardware calibration. Thus the system can conveniently be implemented.

Neuro Fuzzy Logic can be used in the process of selecting the most appropriate tools as well as in Cost Estimation of CNC manufacturing of prototypes, which consists usually in a very small production series, many times as little as a single production unit. In such cases the cutting tools and manufacturing cost must be estimated in a reasonable amount of time and with a certain degree of accuracy, so the shop specialized in this market can be efficient and economically competitive. Therefore an Artificial Intelligence approach can be helpful. Our solution is implemented within an object-oriented programming paradigm for part description and data processing. A set of Fuzzy Logic Rules is applied in a decision-making process to obtain the best tool selection. We use Neural-Networks to generate selforganizing maps (SOM) for adjustment of either the shape and/or boundaries of the membership functions. The optimal cutting conditions are selected by a heuristic approach. Thus the time and cost of a competitive manufacturing job can be adequately predicted.

Motivation: One of the major challenges in the post-genomic era is the speed up of the process of identification of molecular targets related to a specific pathology. Even if the experimental procedure have greatly enhanced the analytical capability, the textual data analysis still play a central role in the planning of the experiments or for database construction. The extraction of relevant information from the published paper requires a lot of time; tools that automatically cluster together the retrieved documents into topic categories labelled by specific relevant keywords can give a great support to this activity. Results: In this paper we present the a application of document clustering system based on Self-Organizing maps to cluster PUBMED abstracts and for the extraction of class specific terms that allow to select the items that are related to some specific topics. The system allows the discrimination of different groups of items and gives an index of relevance for the terms. We have tested the system on a small test sample of PUBMED abstract related to the CDK5 proteins.

Intrusion detection systems monitor computer system events to discover malicious activities in the network. There are two types of intrusion detection systems, namely, signature-based and anomaly-based. Anomaly detection can be either flow-based or packet-based. In the flow-based approach, the system looks at aggregated information of related packets in the form of flow. Packet-based detection system inspects the complete packet which consists of a header as well as payload data. In this paper, a packet-based improved anomaly detection technique is proposed. In the training module, the normal profiles of the network traffic are generated by modeling the payload of the network using n-gram approach by applying lengthwise clustering of packets according to payload length. Lengthwise clustering is done to reduce the number of models for normal profiles. Then the mean and standard deviation is calculated which are used in detection module. In detection module, the distance between normal profiles and newly arriving data in the network is computed using cosine similarity. The standard dataset DARPA'99 and the Panjab University collected data are used for testing the proposed technique. Anomaly detection of the proposed technique is done on port numbers 21, 23 and 80 and the results are compared with the various n-gram techniques and other techniques used in literature for payload anomaly detection. It is concluded that this improved technique can reduce space and provide better results on port 21 and port 23 than on port 80.

Different methodologies are available for clustering purposes. The objective of this paper is to review the capacity of some of them and specifically to test the ability of self-organizing maps (SOMs) to filter, classify, and extract patterns from distributor, commercializer, or customer electrical demand databases. These market participants can achieve an interesting benefit through the knowledge of these patterns, for example, to evaluate the potential for distributed generation, energy efficiency, and demand-side response policies (market analysis). For simplicity, customer classification techniques usually used the historic load curves of each user. The first step in the methodology presented in this paper is anomalous data filtering: holidays, maintenance, and wrong measurements must be removed from the database. Subsequently, two different treatments (frequency and time domain) of demand data were tested to feed SOM maps and evaluate the advantages of each approach. Finally, the ability of SOM to classify new customers in different clusters is also examined. Both steps have been performed through a well-known technique: SOM maps. The results clearly show the suitability of this approach to improve data management and to easily find coherent clusters between electrical users, accounting for relevant information about weekend demand patterns.

We introduce a Self-Organizing Map (SOM) based visualization method that compares cluster structures in temporal datasets using Relative Density SOM (ReDSOM) visualization. Our method, combined with a distance matrix-based visualization, is capable of visually identifying emerging clusters, disappearing clusters, enlarging clusters, contracting clusters, the shifting of cluster centroids, and changes in cluster density. For example, when a region in a SOM becomes significantly more dense compared to an earlier SOM, and well separated from other regions, then the new region can be said to represent a new cluster. The capabilities of ReDSOM are demonstrated using synthetic datasets, as well as real-life datasets from the World Bank and the Australian Taxation Office. The results on the real-life datasets demonstrate that changes identified interactively can be related to actual changes. The identification of such cluster changes is important in many contexts, including the exploration of changes in population behavior in the context of compliance and fraud in taxation.

In some application contexts, data are better described by a matrix of pairwise dissimilarities rather than by a vector representation. Clustering and topographic mapping algorithms have been adapted to this type of data, either via the generalized Median principle, or more recently with the so called relational approach, in which prototypes are represented by virtual linear combinations of the original observations. One drawback of those methods is their complexity, which scales as the square of the number of observations, mainly because they use dense prototype representations: each prototype is obtained as a virtual combination of all the elements of its cluster (at least). We propose in this paper to use a sparse representation of the prototypes to obtain relational algorithms with sub-quadratic complexity.

The self-organizing map (SOM) and neural gas (NG) and generalizations thereof such as the generative topographic map constitute popular algorithms to represent data by means of prototypes arranged on a (hopefully) topology representing map. However, most standard methods rely on the Euclidean metric, hence the resulting clusters are isotropic and they cannot account for local distorsions or correlations of data. In this contribution, we extend prototype-based clustering algorithms such as NG and SOM towards a more general metric which is given by a full adaptive matrix such that ellipsoidal clusters are accounted for. Thereby, the approach relies on a natural extension of the standard cost functions of NG and SOM (in the form of Heskes) and is conceptually intuitive. We derive batch optimization learning rules for prototype and matrix adaptation based on these generalized cost functions and we show convergence of the algorithm. Thereby, it can be seen that matrix learning implicitly performs local principal component analysis (PCA) and the local eigenvectors correspond to the main axes of the ellipsoidal clusters. Thus, the proposal also provides a cost function associated to alternative proposals in the literature which combine SOM or NG with local PCA models. We demonstrate the behavior of the proposed model in several benchmark examples and in an application to image compression.

Clustering constitutes an ubiquitous problem when dealing with huge data sets for data compression, visualization, or preprocessing. Prototype-based neural methods such as neural gas or the self-organizing map offer an intuitive and fast variant which represents data by means of typical representatives, thereby running in linear time. Recently, an extension of these methods towards relational clustering has been proposed which can handle general non-vectorial data characterized by dissimilarities only, such as alignment or general kernels. This extension, relational neural gas, is directly applicable in important domains such as bioinformatics or text clustering. However, it is quadratic in m both in memory and in time (m being the number of data points). Hence, it is infeasible for huge data sets. In this contribution we introduce an approximate patch version of relational neural gas which relies on the same cost function but it dramatically reduces time and memory requirements. It offers a single pass clustering algorithm for huge data sets, running in constant space and linear time only.

Clustering and visualization constitute key issues in computersupported data inspection, and a variety of promising tools exist for such tasks such as the self-organizing map (SOM) and variations thereof. Real life data, however, pose severe problems to standard data inspection: on the one hand, data are often represented by complex non-vectorial objects and standard methods for finite dimensional vectors in Euclidean space cannot be applied. On the other hand, very large data sets have to be dealt with, such that data do neither fit into main memory, nor more than one pass over the data is still affordable, i.e. standard methods can simply not be applied due to the sheer amount of data. We present two recent extensions of topographic mappings: relational clustering, which can deal with general proximity data given by pairwise distances, and patch processing, which can process streaming data of arbitrary size in patches. Together, an efficient linear time data inspection method for general dissimilarity data structures results. We present the theoretical background as well as applications to the areas of text and multimedia processing based on the generalized compression distance.

4.3 Results of matrix NG, SOM, and k-means after 100 epochs for the spirals data set. Obviously, matrix k-means suffers from local optima. 4.4 Results of matrix NG, SOM, and k-means (top) and standard NG, SOM, and k-means (bottom) after 100 epochs for the spirals data set. Obviously, k-means suffers from local optima. Further, the shape is better represented by matrix clustering as can be seen by the classification accuracy of the maps.

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.