Underwater video data are a rich source of information for marine biologists. However, the large amount of recorded video creates a ’big data’ problem, which emphasizes the need for automated detection techniques. This work focuses on the detection of quasi-stationary crabs of various sizes in deep-sea images. Specific issues related to image quality such as low contrast and non-uniform lighting are addressed by the pre-processing step. The segmentation step is based on color, size and shape considerations. Segmentation identifies regions that potentially correspond to crabs. These regions are normalized to be invariant to scale and translation. Feature vectors are formed by the normalized regions, and they are further classified via supervised and non-supervised machine learning techniques. The proposed approach is evaluated experimentally using a video dataset available from Ocean Networks Canada. The thesis provides an in-depth discussion about the performance of the proposed algorithms. / Graduate / 0544 / 0800 / 0547 / mars_mehr@hotmail.com

Current research carried out on financial forecasting has highlighted some limitations of classical econometric methods based on the assumption that the investigated time series can be described as stationary stochastic processes with Gaussian probability density functions. Chaotic behaviour, fractal characteristics and non-linear dynamics have been emerging in different aspects of the financial forecasting problem. The objective of this thesis is to take a system level perspective of the financial forecasting problem and to explore a number of approaches to enhance more 'traditional' decision making flows for stock market forecasting, with particular emphasis on stock selection and timing. To achieve this purpose, a number of stock selection and timing computational 'modules' are investigated. From a computational point of view, the investigation performed in this work encompass techniques such as artificial neural networks, genetic algorithms, chaos theory and fractal geometry, as well as more traditional methods such as clustering, screening, ranking, and statistics based models. From a financial data point of view, this research takes advantage of both fundamental and technical information to enhance the stock selection and timing processes and to cover several investment horizons. Three computational modules are proposed. First, a multivariate stock ranking module which uses fundamental information and is optimised through genetic algorithms. Second, a multivariate forecasting module which uses technical information and is based on artificial neural networks. Third, a univariate price time series forecasting module based on artificial neural networks. In addition, an integrated flow that takes advantage of some synergies and complementary properties of the devised modules is proposed. The effectiveness of the developed modules and the viability of the proposed integrated flow are evaluated over a number of investment horizons using (out-of-sample) historical data.

The friction surfacing process is a derivative of friction welding and retains all the benefits of that welding process - solid phase, forged microstructures and excellent metallurgical bonds. This work is aimed at the development of mathematical and statistical models for the optimisation of the significant process parameters in order to allow rapid development of new applications using standard CNC equipment. Also the possibility of implementing real-time control systems have been investigated and developed. A friction surfacing database has been configured to allow continuos recording and storage of the useful machine outputs. Later, an infrared pyrometer and thermocouples have also been connected to the data acquisition set-up establishing fully automated information flow from the process. A conversion procedure has been developed to ensure that the experimental results are applicable in industrial environments. Response surface map and the method of visual optimisation have been developed. They are an essential part of the methodology for experimental optimisation of the friction surfacing process. The problem of modelling and optimisation has also been approached using accurate statistical methods. Artificial intelligence in the form of neural networks has been used to improve the accuracy of the derived friction surfacing analytical relationships. For the first time dynamic study of the process has been carried out and CARIMA models have been derived using a modified version of the recursive least squares, to ensure high sensitivity and stability of the identification procedure. New conversion technique has been developed, allowing the use of existing models for materials that have not been used for friction surfacing before, reducing significantly the number of experiments. The idea of using indicator parameters has been introduced for the first time in this research. Such parameters are the force, the torque and the interface temperature and they can be measured on-line. It has been shown that variations of these parameters reflect in the quality of the coating characteristics that cannot be measured on-line. Real-time control has also been considered. An algorithm involving fuzzy logic and self-tuning extremum controller has been developed to continuously monitor and compensate in real-time against the variations in the coating characteristics, and respectively in the indicator parameters. The proposed methodology has been used to design a control system that is capable of maintaining optimal process characteristics. The value of this work is also in reducing the lead-time and hence the cost for determining the optimum parameters for a given coating material on a given substrate geometry. This is an important feature when developing new applications for the friction surfacing process. On the basis of this research a range of new commercial applications have emerged including the manufacture of machine knives for the food, pharmaceutical and packaging industries, repair of car engine valve seats, turbine blades, reclamation of shafts, etc.

In financial econometrics the modeling of asset return series is closely related to the estimation of the corresponding conditional densities. One reason why one is interested in the whole conditional density and not only in the conditional mean, is that the conditional variance can be interpreted as a measure of time-dependent volatility of the return series. In fact, the modeling and the prediction of volatility is one of the central topics in asset pricing. In this paper we propose to estimate conditional densities semi-nonparametrically in a neural network framework. Our recurrent mixture density networks realize the basic ideas of prominent GARCH approaches but they are capable of modeling any continuous conditional density also allowing for time-dependent higher-order moments. Our empirical analysis on daily DAX data shows that out-of-sample volatility predictions of the neural network model are superior to predictions of GARCH models in that they have a higher correlation with implied volatilities. (author's abstract) / Series: Report Series SFB "Adaptive Information Systems and Modelling in Economics and Management Science"

This report examines the trends in sick leave usage among nurses in a hospital and aims at creating a forecasting model to predict sick leave usage on a weekly basis using the concept of artificial neural networks (ANN). The data used for the research includes the absenteeism (sick leave) reports for 3 years at a hospital. The analysis shows that there are certain factors that lead to a rise or fall in the weekly sick leave usage. The ANN model tries to capture the effect of these factors and forecasts the sick leave usage for a 1 year horizon based on what it has learned from the behavior of the historical data from the previous 2 years. The various parameters of the model are determined and the model is constructed and tested for its forecasting ability. / text

published_or_final_version / abstract / Physics / Doctoral / Doctor of Philosophy

published_or_final_version / Electrical and Electronic Engineering / Master / Master of Philosophy

No description available.

No description available.

Μια σύντομη αναφορά σε κάποια γνωστά μοντέλα Νευρωνικών Δικτύων, περιγραφή της αρχιτεκτονικής τους και εφαρμογές. Παραδείγματα και εφαρμογές Δυναμικών Νευρωνικών Δικτύων. Γενικό πλαίσιο λειτουργίας των CNN, ιδιότητες και εφαρμογές. / A short reference in Neural Networks, architecture description and applications. Implementation of Dynamic Neural Networks. CNN (cellular neural networks) paradigm, attributes and examples.