Konstantinos Koutroumbas

National Observatory of Athens, Athínai, Attica, Greece

Are you Konstantinos Koutroumbas?

Claim your profile

Publications (21)21.57 Total impact

  • K. Koutroumbas, I. Tsagouri, A. Belehaki
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents an analysis of a set of time series that represent foF2 disturbances during storm conditions, using clustering tools. The time series under study have been drawn from ionospheric observations obtained from eight European middle latitude ionosondes during a significant number of storm-time intervals and they are divided into eight groups according to the latitude (middle to low and middle to high) and the local time of the observation point at storm onset (afternoon, evening, morning, prenoon). The time series in each group have been analyzed using clustering-based methods. Specifically, each time series has been represented using two different ways of representation: the first is the raw representation while the second is through the parameters of the autoregressive (AR) model that best represents it. For each representation a hierarchy of clusterings is produced via the complete link algorithm. The two produced hierarchies are combined to a single one and the final clustering results are extracted from the produced hierarchy. The obtained results are in close agreement with the theoretical formulations concerning ionospheric storm effects at middle latitudes. In addition, they may be proved useful in the development of more accurate ionospheric forecasting methods.
    Advances in Space Research 05/2010; 45(9):1129-1144. · 1.18 Impact Factor
  • Konstantinos Koutroumbas, Yannis Bakopoulos
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper the problem of the approximation of decision regions bordered by (a) closed and/or (b) open and unbounded convex hypersurfaces using feedforward neural networks (FNNs) with hard limiter nodes is considered. Specifically, a constructive proof is given for the fact that a two or a three layer FNN with hard limiter nodes can approximate with arbitrary precision a given decision region of the above kind. This is carried out in three steps. First, each hypersurface is approximated by hyperplanes. Then each one of the regions formed by the hypersurfaces is appropriately approximated by regions defined via the previous hyperplanes. Finally, a feedforward neural network with hard limiter nodes is constructed, based on the previous hyperplanes and the regions defined by them.
    Artificial Intelligence: Theories, Models and Applications, 6th Hellenic Conference on AI, SETN 2010, Athens, Greece, May 4-7, 2010. Proceedings; 01/2010
  • I. Tsagouri, K. Koutroumbas, A. Belehaki
    [Show abstract] [Hide abstract]
    ABSTRACT: The application of the SWIF (Solar Wind driven autoregression model for Ionospheric short term Forecast) model to the European Digital Upper Atmosphere Server - DIAS is discussed in this paper. SWIF is an empirical method designed to provide alerts and warnings for upcoming ionospheric storm disturbances as well as ionospheric forecasts for the middle latitude ionosphere. It combines historical and real-time ionospheric observations with solar wind parameters obtained in real time at L1 point through the cooperation of an autoregression forecasting algorithm, namely TSAR with an empirical ionospheric storm time model, namely STIM that is triggered by solar wind disturbances. SWIF can be considered as a powerful tool for the development of a full set of ionospheric forecasting services and therefore it can efficiently serve the objectives of the DIAS system (http://dias.space.noa.gr). As a first step towards the application of SWIF in DIAS, the work presented here includes the evaluation of SWIF's performance over all DIAS station locations (Athens, Rome, Ebre, Arenosillo, Chilton, Juliusruh, and Pruhonice) and the determination of its prediction accuracy limits for all possible geophysical conditions based on historical data. The results verify the reliability of SWIF's performance for prediction step up to 24 h ahead for all European locations and drive the upgrade of the SWIF algorithm for the accommodation of the prediction error at each prediction step. Finally, the on-line implementation of the SWIF algorithm in DIAS system leads to the substantial upgrade of all DIAS related products and services (e.g. single station forecasts, European forecasting maps, alerts and warnings). This upgrade will strengthen further the basis of the EURIPOS initiative, since reliable ionospheric forecasts are important for the development of new models and tools towards the successful specification of the ionosphere-plasmasphere system.
    03/2009; 11:12293.
  • I. Tsagouri, K. Koutroumbas, A. Belehaki
    [Show abstract] [Hide abstract]
    ABSTRACT: The development of a new ionospheric forecasting algorithm, called the Solar Wind driven autoregression model for Ionospheric short-term Forecast (SWIF) is presented. SWIF combines historical and real-time ionospheric observations with solar wind parameters obtained in real time at the L1 point. This is achieved through the cooperation of an autoregression forecasting algorithm, called Time Series AutoRegressive (TSAR), with the empirical Storm Time Ionospheric Model that formulates the ionospheric storm time response based on solar wind input. The evaluation of SWIF's predictions was principally focused on its performance during selected storm time intervals over Europe. The results demonstrate significant improvement of SWIF's prediction capability over its predecessor, TSAR, significant improvement over climatology and evidence of SWIF's efficiency compared to other forecasting methods. Moreover, the evaluation of SWIF's output over long time periods, that include a wide range of geophysical conditions, suggests that SWIF can be used for prediction up to 24 h ahead.
    Radio Science 02/2009; 44(1). · 1.00 Impact Factor
  • Konstantinos Koutroumbas
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper the problem of the determination of the maximum among the M members of a set of positive real numbers S is considered. More specifically, a version of the Hamming Maxnet is proposed that is able to determine all maxima of S, in contrast to the original Hamming Maxnet and most of its variants, which can not deal with multiple maxima in S. A detailed convergence analysis of the proposed network is provided. Also, the proposed version is compared with other variants of the Hamming Maxnet via simulations.
    09/2008: pages 135-147;
  • Source
    Koutroumbas K, Tsagouri I, Belehaki A
    [Show abstract] [Hide abstract]
    ABSTRACT: A new method for ionospheric predictions based on time series autoregressive models (AR) that was recently developed to serve the needs of the European Digital Upper Atmosphere Server (DIAS) for short term forecast of the foF2 parameter over Europe (up to the next 24 h) is described. Its performance for various steps ahead is compared with the outcome of neural network predictors for both storm and quiet periods in two DIAS locations, Athens and Pruhonice. The results indicate that the proposed method provides robust short term forecasts of the foF2 for the middle latitude ionosphere.
    Annales Geophysicae 01/2008; · 1.52 Impact Factor
  • Nicholas Kalouptsidis, Konstantinos Koutroumbas, V. Psaraki
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper random utility maximization based on maximization of correct classification of the choice decisions over a given data set is considered. It is shown that if the disturbance vector in the random utility model is independent and identically distributed, then preference determination based on the most probable alternative reduces to deterministic utility maximization. As a consequence of the above equivalence, the form of the error distribution (normal, Weibull, uniform etc.) plays no role in the determination of the preferred alternative. Parameter estimation under the most probable alternative rule is carried out using two methods. The first is based on the solution of an appropriately defined system of linear inequalities and the second one is based on the function optimization of a newly proposed function, whose optimum is achieved when the number of correctly classified individuals is maximized. The ability to use these algorithms in the framework of pattern recognition and machine learning is pointed out. Simulations and a real case study involving intercity travel behavior are employed to assess the proposed methods.
    European Journal of Operational Research 01/2007; 176:1778-1794. · 2.04 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: The objective of this study was the investigation of the potential value of morphometry, feature selection and statistical classifiers techniques, such as neural networks, for the classification of benign from malignant gastric nuclei and cases. One hundred and twenty gastric smears, routinely processed and stained by Papanicolaou technique, were analyzed by a customized image analysis system. Data from half of the cases were selected to form the training set, while the remaining data formed the test set. A feature selection technique was applied in order to identify the most important nuclear features, which were used in a second stage by statistical classifiers to classify a nucleus as benign or malignant. Using the classifier results for the nuclear classification, a method to classify each individual patient was developed. The performance of the proposed method was validated through the test set. The technique described in this report produces significant results at the nuclear and patient level and promises to be a powerful assistance tool for everyday cytological laboratory routine.
    Oncology Reports 02/2006; 15 Spec no.:1033-6. · 2.30 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: The objective of this study was to perform a comparative investigation of the capability of various classifiers in discriminating benign from malignant thyroid lesions. Using May Grunvald-Giemsa-stained smears taken by fine needle aspiration (FNA) and a custom image analysis system, 25 nuclear features describing the size, shape and texture of the nuclei were measured in each case. A statistical pre-processing of features revealed that only 4 of the 25 features are important when discriminating benign from malignant thyroid lesions, which were transformed and fed to four classifiers for subsequent analysis. The cases were divided into one set used for the training of classifiers, a second set used as the test set, and the remaining cases with no clear classification formed an ambiguous test set. Classification was performed at the nuclear and patient level. The technique described in this study produced encouraging results and promises to be a helpful tool in the daily cytological laboratory routine.
    Oncology Reports 02/2006; 15 Spec no.:1023-6. · 2.30 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: The objective of this study is to investigate the discrimination of benign from malignant breast lesions using: the linear, the feedforward neural network, the k-nearest neighbor and the boosting classifiers. Nuclear morphometric parameters from cytological smears taken by Fine Needle Aspiration (FNA) of the breast, have been measured from 193 patients. These parameters undergo an appropriate transformation and then, the classifiers are performed on the raw and on the transformed data. The results show that in terms of the raw data set all classifiers exhibit almost the same performance (overall accuracy ≡ 87%), Thus the linear classifier suffices for the discrimination of the present problem. Also, based on the previous results, one can conjecture that the use of these classifiers combined with image morphometry and statistical techniques for feature transformation, may offer useful information towards the improvement of the diagnostic accuracy of breast FNA.
    Advances in Artificial Intelligence, 4th Helenic Conference on AI, SETN 2006, Heraklion, Crete, Greece, May 18-20, 2006, Proceedings; 01/2006
  • Konstantinos Koutroumbas, Nicholas Kalouptsidis
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper the classical Hamming network is generalized in various ways. First, for the Hamming maxnet, a generalized model is proposed, which covers under its umbrella most of the existing versions of the Hamming Maxnet. The network dynamics are time varying while the commonly used ramp function may be replaced by a much more general non-linear function. Also, the weight parameters of the network are time varying. A detailed convergence analysis is provided. A bound on the number of iterations required for convergence is derived and its distribution functions are given for the cases where the initial values of the nodes of the Hamming maxnet stem from the uniform and the peak distributions. Stabilization mechanisms aiming to prevent the node(s) with the maximum initial value diverging to infinity or decaying to zero are described. Simulations demonstrate the advantages of the proposed extension. Also, a rough comparison between the proposed generalized scheme as well as the original Hamming maxnet and its variants is carried out in terms of the time required for convergence, in hardware implementations. Finally, the other two parts of the Hamming network, namely the competitors generating module and the decoding module, are briefly considered in the framework of various applications such as classification/clustering, vector quantization and function optimization.
    Neural Networks 10/2005; 18(7):896-913. · 1.93 Impact Factor
  • K. Koutroumbas, A. Belehaki
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper the problem of one-step ahead prediction of the critical frequency ( fo F2) of the middle-latitude ionosphere, using time series forecasting methods, is considered. The whole study is based on a sample of about 58000 observations of fo F2 with 15-min time resolution, derived from the Athens digisonde ionograms taken from the Digisonde Portable Sounder (DPS4) located at Palaia Penteli (38° N, 23.5° E), for the period from October 2002 to May 2004. First, the embedding dimension of the dynamical system that generates the above sample is estimated using the false nearest neighbor method. This information is then utilized for the training of the predictors employed in this study, which are the linear predictor, the neural network predictor, the persistence predictor and the k -nearest neighbor predictor. The results obtained by the above predictors suggest that, as far as the mean square error is considered as performance criterion, the first two predictors are significantly better than the latter two predictors. In addition, the results obtained by the linear and the neural network predictors are not significantly different from each other. This may be taken as an indication that a linear model suffices for one step ahead prediction of fo F2.
    Annales Geophysicae 01/2005; · 1.52 Impact Factor
  • Source
    Konstantinos Koutroumbas
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, a novel method for the determination of the position of the maxima among the M members of a set of positive real numbers S that stem from a finite discrete domain is proposed. The method does not follow the competitive spirit of the winner-take-all methods. It involves neither recursion nor comparisons between the members of S. Instead, a threshold T less than but arbitrarily close to the maximum value of S is directly calculated, with the contribution of all members of S and then each member of S is compared with it. Also, it is shown how the proposed method can be applied to the associative memory problem. In addition, arguments are given showing that, in principle, there are versions of the proposed method that are significantly faster than other well established methods for determining the position of the maximum in a set S of numbers.
    Neural Processing Letters 01/2005; 22:205-221. · 1.24 Impact Factor
  • Source
    Konstantinos Koutroumbas
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, two novel recurrent algorithms for selecting the maxima of a set S containing M positive real numbers are introduced. In the first one the aim is to determine a threshold T such that only the maxima of S lie above it, while in the second one, each element of S is reduced independently of the rest until either it becomes zero (if it is non-maximum) or it freezes to a positive value (if it is a maximum). Convergence analysis of both schemes as well as neural network implementations of simplified versions of the algorithms are given. Finally, a comparison of the performance of the proposed algorithms with other related methods is carried out.
    Neural Processing Letters 01/2004; 20:179-197. · 1.24 Impact Factor
  • K Koutroumbas
    [Show abstract] [Hide abstract]
    ABSTRACT: In this letter, the capabilities of feedforward neural networks (FNNs) on the realization and approximation of functions of the form g: R(l) --> A, which partition the R(l) space into polyhedral sets, each one being assigned to one out of the c classes of A, are investigated. More specifically, a constructive proof is given for the fact that FNNs consisting of nodes having sigmoid output functions are capable of approximating any function g with arbitrary accuracy. Also, the capabilities of FNNs consisting of nodes having the hard limiter as output function are reviewed. In both cases, the two-class as well as the multiclass cases are considered.
    Neural Computation 10/2003; 15(10):2457-81. · 1.76 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: To investigate the potential value of morphometry and neural networks for the discrimination of benign from malignant gastric lesions. One thousand cells from 19 cases of cancer, 19 cases of gastritis and 56 cases of ulcer were selected as a training set, and an additional 4,000 cells from the same cases of cancer, gastritis and ulcer were used as a test set. Images of routinely processed gastric smears stained by the Papanicolaou technique were analyzed by a custom-made image analysis system. Application of the neural network gave correct classification in 96% of benign cells and 89% of malignant cells. The results indicate that the use of neural networks and image morphometry may offer useful information concerning the potential of malignancy in gastric cells.
    Analytical and quantitative cytology and histology / the International Academy of Cytology [and] American Society of Cytology 03/2000; 22(1):63-9. · 0.60 Impact Factor
  • K. Koutroumbas, N. Kalouptsidis
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper two neural network architectures for selecting the maximum among a set of numbers are introduced. The first architecture is recurrent and relies on the Hamming MaxNet. The second architecture is feedforward, featuring modularity and pipelineability.
    International Journal of Computer Mathematics - IJCM. 01/1998; 67:25-32.
  • K. Koutroumbas, N. Kalouptsidis
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper convergence analysis of the parallel and deterministic asynchronous modes of operation for the Hamming network is carried out. Conditions ensuring convergence to a stable state in a finite number of steps are derived. An upper bound of the maximum number of steps that is required to reach a stable state is obtained. Finally, a geometrical interpretation of our results is obtained
    IEEE Transactions on Neural Networks 06/1994; · 2.95 Impact Factor
  • K. Koutroumbas, N. Kalouptsidis
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper two algorithms for the construction of pattern classifier neural architectures are proposed. A comparison with other known similar architectures is given and simulation results are carried out
    Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on; 01/1994
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we perform a comparative evaluation of four different computational learning methods on a problem of diagnostic cytology and more specifically on the classification of gastric cells. The methods considered are: Decision Tree Induction, Boosted Decision Trees, Naive Bayesian Classifier, and Radial Basis Function Neural Networks. The performance of each method was assessed on unseen data. Our aim was not to evaluate the quality of the algorithms as such, but to examine which of them are suitable for the specific medical diagnosis task, in order to provide a reliable diagnostic tool to the doctors involved in the area. We compare the performance of the four methods and discuss the results taking into account the characteristics of the methods and the task examined. The dataset that was used in this paper is publicly available, facilitating reproducibility of the results and providing a basis of comparison for future work.