Panos Liatsis

Dynamic Ridge Polynomial Neural Network: Forecasting the univariate non-stationary and stationary trading signals

This paper considers the prediction of noisy time series data, specifically, the prediction of financial signals. A novel Dynamic Ridge Polynomial Neural Network (DRPNN) for financial time series prediction is presented which combines the properties of both higher order and recurrent neural network. In an attempt to overcome the stability and convergence problems in the proposed DRPNN, the stability convergence of DRPNN is derived to ensure that the network posses a unique equilibrium state. In order to provide a more accurate comparative evaluation in terms of profit earning, empirical testing used in this work encompass not only on the more traditional criteria of NMSE, which concerned at how good the forecasts fit their target, but also on financial metrics where the objective is to use the networks predictions to generate profit. Extensive simulations for the prediction of one and five steps ahead of stationary and non-stationary time series were performed. The resulting forecast made by DRPNN shows substantial profits on financial historical signals when compared to various neural networks; the Pi-Sigma Neural Network, the Functional Link Neural Network, the feedforward Ridge Polynomial Neural Network, and the Multilayer Perceptron. Simulation results indicate that DRPNN in most cases demonstrated advantages in capturing chaotic movement in the financial signals with an improvement in the profit return and rapid convergence over other network models.

A new method of three‐dimensional coronary artery reconstruction from X‐ray angiography: Validation against a virtual phantom and multislice computed tomography

Objective: To develop and implement a method for three‐dimensional (3D) reconstruction of coronary arteries from conventional monoplane angiograms. Background: 3D reconstruction of conventional coronary angiograms is a promising imaging modality for both diagnostic and interventional purposes. Methods: Our method combines image enhancement, automatic edge detection, an iterative method to reconstruct the centerline of the artery and reconstruction of the diameter of the vessel by taking into consideration foreshortening effects. The X‐Ray‐based 3D coronary trees were compared against phantom data from a virtual arterial tree projected into two planes as well as computed tomography (CT)‐based coronary artery reconstructions in patients subjected to coronary angiography. Results: Comparison against the phantom arterial tree demonstrated perfect agreement with the developed algorithm. Visual comparison against the CT‐based reconstruction was performed in the 3D space, in terms of the direction angle along the centerline length of the left anterior descending and circumflex arteries relative to the main stem, and location and take‐off angle of sample bifurcation branches from the main coronary arteries. Only minimal differences were detected between the two methods. Inter‐ and intraobserver variability of our method was low (intra‐class correlation coefficients > 0.8). Conclusion: The developed method for coronary artery reconstruction from conventional angiography images provides the geometry of coronary arteries in the 3D space.

Appearance-based statistical methods for face recognition

Different statistical methods for face recognition have been proposed in recent years. They mostly differ in the type of projection and distance measure used. The aim of this paper is to give an overview of most popular statistical subspace methods for face recognition task. Theoretical aspects of three algorithms will be considered and some reported performance evaluations will be given.

Simulation of cardiac motion on non-Newtonian, pulsating flow development in the human left anterior descending coronary artery.

This study aimed at investigating the effect of myocardial motion on pulsating blood flow distribution of the left anterior descending coronary artery in the presence of atheromatous stenosis. The moving 3D arterial tree geometry has been obtained from conventional x-ray angiograms obtained during the heart cycle and includes a number of major branches. The geometry reconstruction model has been validated against projection data from a virtual phantom arterial tree as well as with CT-based reconstruction data for the same patient investigated. Reconstructions have been obtained for a number of temporal points while linear interpolation has been used for all intermediate instances. Blood has been considered as a non-Newtonian fluid. Results have been obtained using the same pulse for the inlet blood flow rate but with fixed arterial tree geometry as well as under steady-state conditions corresponding to the mean flow rate. Predictions indicate that myocardial motion has only a minor effect on flow distribution within the arterial tree relative to the effect of the blood pressure pulse.

Density-Driven Generalized Regression Neural Networks (DD-GRNN) for Function Approximation

This paper proposes a new nonparametric regression method, based on the combination of generalized regression neural networks (GRNNs), density-dependent multiple kernel bandwidths, and regularization. The presented model is generic and substitutes the very large number of bandwidths with a much smaller number of trainable weights that control the regression model. It depends on sets of extracted data density features which reflect the density properties and distribution irregularities of the training data sets. We provide an efficient initialization scheme and a second-order algorithm to train the model, as well as an overfitting control mechanism based on Bayesian regularization. Numerical results show that the proposed network manages to reduce significantly the computational demands of having individual bandwidths, while at the same time, provides competitive function approximation accuracy in relation to existing methods.

Development of a neonate lung reconstruction algorithm using a wavelet AMG and estimated boundary form

Objective, non-invasive measures of lung maturity and development, oxygen requirements and lung function, suitable for use in small, unsedated infants, are urgently required to define the nature and severity of persisting lung disease, and to identify risk factors for developing chronic lung problems. Disorders of lung growth, maturation and control of breathing are among the most important problems faced by the neonatologists. At present, no system for continuous monitoring of neonate lung function to reduce the risk of chronic lung disease in infancy in intensive care units exists. We are in the process of developing a new integrated electrical impedance tomography (EIT) system based on wearable technology to integrate measures of the boundary diameter from the boundary form for neonates into the reconstruction algorithm. In principle, this approach could provide a reduction of image artefacts in the reconstructed image associated with incorrect boundary form assumptions. In this paper, we investigate the required accuracy of the boundary form that would be suitable to minimize artefacts in the reconstruction for neonate lung function. The number of data points needed to create the required boundary form is automatically determined using genetic algorithms. The approach presented in this paper is to assist quality of the reconstruction using different approximations to the ideal boundary form. We also investigate the use of a wavelet algebraic multi-grid (WAMG) preconditioner to reduce the reconstruction computation requirements. Results are presented that demonstrate a full 3D model is required to minimize artefact in the reconstructed image and the implementation of a WAMG for EIT.

A GMDH-based fuzzy modeling approach for constructing TS model

In this paper, a new learning algorithm based on group method of data handling (GMDH) is proposed for the identification of Takagi-Sugeno fuzzy model. Different from existing methods, the new approach, called TS-GMDH, starts from simple elementary TS fuzzy models, and then uses the mechanism of GMDH to produce candidate fuzzy models of growing complexity until the TS model of optimal complexity has been created. The main characteristic of the new approach is its ability to identify the structure of TS model automatically. Experiments on Box-Jenkins gas furnace data and UCI datasets have shown that the proposed method can achieve satisfactory results and is more robust to noise in comparison with other TS modeling techniques such as ANFIS.

The application of ridge polynomial neural network to multi-step ahead financial time series prediction

Motivated by the slow learning properties of multilayer perceptrons (MLPs) which utilize computationally intensive training algorithms, such as the backpropagation learning algorithm, and can get trapped in local minima, this work deals with ridge polynomial neural networks (RPNN), which maintain fast learning properties and powerful mapping capabilities of single layer high order neural networks. The RPNN is constructed from a number of increasing orders of Pi–Sigma units, which are used to capture the underlying patterns in financial time series signals and to predict future trends in the financial market. In particular, this paper systematically investigates a method of pre-processing the financial signals in order to reduce the influence of their trends. The performance of the networks is benchmarked against the performance of MLPs, functional link neural networks (FLNN), and Pi–Sigma neural networks (PSNN). Simulation results clearly demonstrate that RPNNs generate higher profit returns with fast convergence on various noisy financial signals.

A robust missing value imputation method for noisy data

Missing data imputation is an important research topic in data mining. The impact of noise is seldom considered in previous works while real-world data often contain much noise. In this paper, we systematically investigate the impact of noise on imputation methods and propose a new imputation approach by introducing the mechanism of Group Method of Data Handling (GMDH) to deal with incomplete data with noise. The performance of four commonly used imputation methods is compared with ours, called RIBG (robust imputation based on GMDH), on nine benchmark datasets. The experimental result demonstrates that noise has a great impact on the effectiveness of imputation techniques and our method RIBG is more robust to noise than the other four imputation methods used as benchmark.

Generalized Regression Neural Networks With Multiple-Bandwidth Sharing and Hybrid Optimization

This paper proposes a novel algorithm for function approximation that extends the standard generalized regression neural network. Instead of a single bandwidth for all the kernels, we employ a multiple-bandwidth configuration. However, unlike previous works that use clustering of the training data for the reduction of the number of bandwidths, we propose a distinct scheme that manages a dramatic bandwidth reduction while preserving the required model complexity. In this scheme, the algorithm partitions the training patterns to groups, where all patterns within each group share the same bandwidth. Grouping relies on the analysis of the local nearest neighbor distance information around the patterns and the principal component analysis with fuzzy clustering. Furthermore, we use a hybrid optimization procedure combining a very efficient variant of the particle swarm optimizer and a quasi-Newton method for global optimization and locally optimal fine-tuning of the network bandwidths. Training is based on the minimization of a flexible adaptation of the leave-one-out validation error that enhances the network generalization. We test the proposed algorithm with real and synthetic datasets, and results show that it exhibits competitive regression performance compared to other techniques.