In this thesis I look into how one can train and run an artificial neural network using Compute Shader and what kind of performance can be expected. An artificial neural network is a computational model that is inspired by biological neural networks, e.g. a brain. Finding what kind of performance can be expected was done by creating an implementation that uses Compute Shader and then compare it to the FANN library, i.e. a fast artificial neural network library written in C. The conclusion is that you can improve performance by training an artificial neural network on the compute shader as long as you are using non-trivial datasets and neural network configurations.

Yacht races are won by good sailors racing fast boats. A good skipper takes decisions at key moments of the race based on the anticipated wind behaviour and on his position on the racing area and with respect to the competitors. His aim is generally to complete the race before all his opponents, or, when this is not possible, to perform better than some of them. In the past two decades some methods have been proposed to compute optimal strategies for a yacht race. Those strategies are aimed at minimizing the expected time needed to complete the race and are based on the assumption that the faster a yacht, the higher the number of races that it will win (and opponents that it will defeat). In a match race, however, only two yachts are competing. A skipper’s aim is therefore to complete the race before his opponent rather than completing the race in the shortest possible time. This means that being on average faster may not necessarily mean winning the majority of races. This thesis sets out to investigate the possibility of computing a sailing strategy for a match race that can defeat an opponent who is following a fixed strategy that minimises the expected time of completion of the race. The proposed method includes two novel aspects in the strategy computation: A short-term wind forecast, based on an Artificial Neural Network (ANN) model, is performed in real time during the race using the wind measurements collected on board. Depending on the relative position with respect to the opponent, decisions with different levels of risk aversion are computed. The risk attitude is modeled using Coherent Risk Measures. The proposed algorithm is implemented in a computer program and is tested by simulating match races between identical boats following progressively refined strategies. Results presented in this thesis show how the intuitive idea of taking more risk when losing and having a conservative attitude when winning is confirmed in the risk model used. The performance of ANN for short-term wind forecasting is tested both on wind speed and wind direction. It is shown that for time steps of the order of seconds and adequate computational power ANN perform better than linear models (persistence models, ARMA) and other nonlinear models (Support Vector Machines). The outcome of the simulated races confirms that maximising the probability of winning a match race does not necessarily correspond to minimising the expected time needed to complete the race.

The aim of thesis is the recognition of the symbols in musical notation. Functions are implemented searching for a template in the image.

Despite the impressive advancements in people detection and tracking, safety is still a key barrier to the deployment of autonomous vehicles in urban environments [1]. For example, in non-autonomous technology, there is an implicit communication between the people crossing the street and the driver to make sure they have communicated their intent to the driver. Therefore, it is crucial for the autonomous car to infer the future intent of the pedestrian quickly. We believe that human body orientation with respect to the camera can help the intelligent unit of the car to anticipate the future movement of the pedestrians. To further improve the safety of pedestrians, it is important to recognize whether they are distracted, carrying a baby, or pushing a shopping cart. Therefore, estimating the fine- grained 3D pose, i.e. (x,y,z)-coordinates of the body joints provides additional information for decision-making units of driverless cars. In this dissertation, we have proposed a deep learning-based solution to classify the categorized body orientation in still images. We have also proposed an efficient framework based on our body orientation classification scheme to estimate human 3D pose in monocular RGB images. Furthermore, we have utilized the dynamics of human motion to infer the body orientation in image sequences. To achieve this, we employ a recurrent neural network model to estimate continuous body orientation from the trajectories of body joints in the image plane. The proposed body orientation and 3D pose estimation framework are tested on the largest 3D pose estimation benchmark, Human3.6m (both in still images and video), and we have proved the efficacy of our approach by benchmarking it against the state-of-the-art approaches. Another critical feature of self-driving car is to avoid an obstacle. In the current prototypes the car either stops or changes its lane even if it causes other traffic disruptions. However, there are situations when it is preferable to collide with the object, for example a foam box, rather than take an action that could result in a much more serious accident than collision with the object. In this dissertation, for the first time, we have presented a novel method to discriminate between physical properties of these types of objects such as bounciness, elasticity, etc. based on their motion characteristics . The proposed algorithm is tested on synthetic data, and, as a proof of concept, its effectiveness on a limited set of real-world data is demonstrated.

Language models play an important role in many natural language processing tasks. In this thesis, we focus on language models built on artificial neural net- works. We examine the possibilities of using morphological annotations in these models. We propose a neural network architecture for a language model that explicitly makes use of morphological annotation of the input sentence: instead of word forms it processes lemmata and morphological tags. Both the baseline and the proposed method are evaluated on their own by perplexity, and also in the context of machine translation by the means of automatic translation quality evaluation. While in isolation the proposed model significantly outperforms the baseline, there is no apparent gain in machine translation. 1

Abstract Drying, especially rotary drying, is without doubt one of the oldest and most common unit operations in the process industries. Rotary dryers are workhorses which are easy and reliable to operate, but neither energy-efficient nor environmentally friendly. In order to conform better to the requirements of modern society concerning working conditions, safety practices and environmental aspects, the development of control systems can provide opportunities for improving dryer operation and efficiency. Our in depth understanding of rotary drying is poor, because it is a very complex process that includes the movement of solids in addition to thermal drying. Thus even today rotary dryers are controlled partly manually, based on the operator's "eye" and experience, and partly relying on conventional control methods. The control of a rotary dryer is difficult due to the long time delay, which means that accidental variations in the input variables can disturb the process for long periods of time before they are reflected in the output variables. To eliminate such disturbances at an early stage, increasing interest has been shown in more sophisticated control systems such as model-based constructs, fuzzy logic and neural nets in recent years. Although it has proved difficult and time-consuming to develop model-based control systems, due to the complexity of the process, intelligent control methods based on fuzzy logic and neural nets offer attractive solutions for improving dryer control. These methods make it possible to utilize experience, knowledge and historical data, large amounts of which are readily available. The aim of this research was to improve dryer control by developing new hybrid control systems, one consisting of a fuzzy logic controller (FLC) and PI controller and the other of a three-layer neural network (NN) and PI controller. The FLC and NN act as supervisory controllers giving set points for the PI controllers. The performance of each was examined both with simulations and in pilot plant experiments. The pilot plant dryer at the University of Oulu closely resembles a real industrial situation, so that the results are relevant. Evaluation of these results showed that the intelligent hybrid controllers are well suited for the control of a rotary dryer, giving a performance in which disturbances can be eliminated rapidly and operation of the dryer can thereby be improved, with the aim of enhancing its efficiency and environmental friendliness.

It is estimated that 65 million people worldwide have epilepsy, and many of them have uncontrollable seizures even with the use of medication. A seizure occurs when the normal electrical activity of the brain is interrupted by sudden and unusually intense bursts of electrical energy, and these bursts can be observed and detected by the use of an electroencephalograph (EEG) machine. This work presents an algorithm that monitors subtle changes in scalp EEG characteristics to predict seizures. The algorithm is built to calibrate itself to every specifc patient based on recorded data, and is computationally effcient enough for future on-line applications. The presented algorithm performs ICA-based artifact filtering and Lasso-based feature selection from a large array of statistical features. Classification is based on a neural network using Bayesian regularized backpropagation.The selected method was able to classify 4 second long preictal segments with an average sensitivity of 99.53% and an average specificity of 99.9% when tested on 15 different patients from the CHB-MIT database.

Many scientific, engineering and economic problems involve the optimisation of a set of parameters. These problems include examples like minimising the losses in a power grid by finding the optimal configuration of the components, or training a neural network to recognise images of people's faces. Numerous optimisation algorithms have been proposed to solve these problems, with varying degrees of success. The Particle Swarm Optimiser (PSO) is a relatively new technique that has been empirically shown to perform well on many of these optimisation problems. This thesis presents a theoretical model that can be used to describe the long-term behaviour of the algorithm. An enhanced version of the Particle Swarm Optimiser is constructed and shown to have guaranteed convergence on local minima. This algorithm is extended further, resulting in an algorithm with guaranteed convergence on global minima. A model for constructing cooperative PSO algorithms is developed, resulting in the introduction of two new PSO-based algorithms. Empirical results are presented to support the theoretical properties predicted by the various models, using synthetic benchmark functions to investigate specific properties. The various PSO-based algorithms are then applied to the task of training neural networks, corroborating the results obtained on the synthetic benchmark functions. / Thesis (PhD)--University of Pretoria, 2007. / Computer Science / Unrestricted

An automated detection process for Eukaryotic ribonucleic acid (RNA) Polymerase II Promoter is presented in this dissertation. We employ an artificial neural network (ANN) in conjunction with features that were selected using an information-theoretic approach. Firstly an introduction is given where the problem is described briefly. Some background is given about the biological and genetic principles involved in DNA, RNA and Promoter detection. The automation process is described with each step given in detail. This includes the data information gathering, feature generation, and the full ANN process. The ANN section of the project is split up in a generation process, a training section as well as a testing section. Lastly the final detection program was tested and compared to other promoter detection systems. An improvement of at least 10% in positive prediction value (PPV) in comparison with current state-of-the-art solutions was obtained. Note: A Companion CD should accompany this report that contains all the program code and some of the source data that was used in this project. All the references to “Companion CD”, reference number [18] are references to these programs.acquisition process, how the different samples were split into different sets and statistical. / Dissertation (MEng (Computer Engineering))--University of Pretoria, 2007. / Electrical, Electronic and Computer Engineering / unrestricted

Feature engineering is a process that augments the feature vector of a machine learning model with calculated values that are designed to enhance the accuracy of a model’s predictions. Research has shown that the accuracy of models such as deep neural networks, support vector machines, and tree/forest-based algorithms sometimes benefit from feature engineering. Expressions that combine one or more of the original features usually create these engineered features. The choice of the exact structure of an engineered feature is dependent on the type of machine learning model in use. Previous research demonstrated that various model families benefit from different types of engineered feature. Random forests, gradient-boosting machines, or other tree-based models might not see the same accuracy gain that an engineered feature allowed neural networks, generalized linear models, or other dot-product based models to achieve on the same data set. This dissertation presents a genetic programming-based algorithm that automatically engineers features that increase the accuracy of deep neural networks for some data sets. For a genetic programming algorithm to be effective, it must prioritize the search space and efficiently evaluate what it finds. This dissertation algorithm faced a potential search space composed of all possible mathematical combinations of the original feature vector. Five experiments were designed to guide the search process to efficiently evolve good engineered features. The result of this dissertation is an automated feature engineering (AFE) algorithm that is computationally efficient, even though a neural network is used to evaluate each candidate feature. This approach gave the algorithm a greater opportunity to specifically target deep neural networks in its search for engineered features that improve accuracy. Finally, a sixth experiment empirically demonstrated the degree to which this algorithm improved the accuracy of neural networks on data sets augmented by the algorithm’s engineered features.