No description available.

This thesis describes the implementation of a Radial Basis Function (RBF) network to be used in predicting the effectiveness of various strategies for reducing the Vehicle Trip Rate (VTR) of a worksite. Three methods of learning were utilized in training the Gaussian hidden units of the network, those being a) output weight adjustment using the Delta rule b) adjustable reference vectors in conjunction with weight adjustment, and c) a combination of adjustable centers and adjustable sigma values for each RBF neuron with the Delta rule. The justification for utilizing each of the more advanced levels of training is provided using a series of tests and performance comparisons. The network architecture is then selected based upon a series of initial trials for an optimum number of hidden Radial Basis neurons. In a similar manner, the training time is determined after finding a maximum number of epochs during which there is a significant change in the SSE. The network was compared for effectiveness against each of the following methods of data analysis: force-entered regression, backward regression, forward regression, stepwise regression, and two types of back-propagation networks based upon the attributes discovered to be most predictive by these regression techniques. A comparison of the learning methods used on the Radial Basis network shows the third learning strategy to be the most efficient for training, yielding the lowest sum of squared errors (SSE) in the shortest number of training epochs. The result of comparing the RBF implementation against the other methods mentions shows the superiority of the Radial Basis method for predictive ability.

Graphics processing units (GPUs) were originally used solely for the purpose of graph- ics rendering. This changed with the introduction of technologies like CUDA that enabled to use graphics processors as any other computing device. However, writing an efficient program for GPUs, also called GPU kernel, is one of the most difficult programming disciplines. The latest research in the field suggests that these difficulties could be po- tentially mitigated with machine learning methods. One especially successful approach is based on the utilization of recurrent neural networks (RNNs) over different representa- tions of source code. In this work, we present two RNN-based solutions that are able to derive performance characteristics of a CUDA GPU kernel directly from its intermediate representation called PTX. We assess the applicability of our two methods in two GPU op- timization tasks. In heterogeneous device mapping task, our methods are able to achieve accuracies of around 82%, results that are slightly worse than the current state of the art. In a more challenging achieved occupancy task, where the goal is to correctly predict one out of ten classes, our two methods achieve accuracies above 50%. These promising results indicate great potential in additional research focused in a similar direction. 1

<p>Electric load forecasting is a critical tool for power system planning and the creation of sustainable energy systems. Precise and reliable load forecasting enables power system operators to make informed decisions regarding power generation and transmission, optimize energy efficiency, and reduce operational costs and extra power generation costs, to further reduce environment-related issues. However, achieving desirable forecasting performance remains challenging due to the irregular, nonstationary, nonlinear, and noisy nature of the observed data under unprecedented events. In recent years, deep learning and other artificial intelligence techniques have emerged as promising approaches for load forecasting. These techniques have the ability to capture complex patterns and relationships in the data and adapt to changing conditions, thereby enhancing forecasting accuracy. As such, the use of deep learning and other artificial intelligence techniques in load forecasting has become an increasingly popular research topic in the field of power systems. </p> <p>Although deep learning techniques have advanced load forecasting, the field still requires more accurate and efficient models. One promising approach is federated learning, which allows for distributed data analysis without exchanging data among multiple devices or cen- ters. This method is particularly relevant for load forecasting, where each power station’s data is sensitive and must be protected. In this study, a proposed approach utilizing Federated Deeponet for seven different power stations is introduced, which proposes a Federated Deep Operator Networks and a Lagevin Dynamics-based Federated Deep Operator Networks using Stochastic Gradient Langevin Dynamics as the optimizer for training the data daily for one-day and predicting for one-day frequencies by frequencies. The data evaluation methods include mean absolute percentage error and the percentage coverage under confidence interval. The findings demonstrate the potential of federated learning for secure and precise load forecasting, while also highlighting the challenges and opportunities of implementing this approach in real-world scenarios. </p>

This Master's Thesis is focused on the principles of neural networks, primarily convolutional neural networks (CNN). It introduces the evolutionary optimization in the context of neural networks. One of existing libraries devoted to the CNN design was chosen (Keras), analysed and used in image classification tasks. An optimization technique based on cartesian genetic programming that should reduce the complexity of CNN's computation was proposed and implemented. The impact of the proposed technique on CNN behaviour was evaluated in a case study.

Quantum computing is a form of computation that exploits quantum mechanical phenomena for information processing, with promising applications (among others) in optimization and machine learning. Indeed, quantum machine learning is currently one of the most popular directions of research in quantum computing, offering solutions with an at-least-theoretical advantage compared to the classical counterparts. Nevertheless, the quantum devices available in the current Noisy Intermediate-Scale Quantum (NISQ) era are limited in the number of qubits and significantly affected by noise. An interesting alternative to the current prototypes of general-purpose quantum devices is represented by quantum annealers, specific-purpose quantum machines implementing the heuristic search for solving optimization problems known as quantum annealing. However, despite the higher number of qubits, the current quantum annealers are characterised by very sparse topologies. These practical issues have led to the development of hybrid classical-quantum schemes, aiming at leveraging the strengths of both paradigms while circumventing some of the limitations of the available devices. In this thesis, several hybrid classical-quantum algorithms for optimization and machine learning are introduced and/or empirically assessed, as the empirical evaluation is a fundamental part of algorithmic research. The quantum computing models taken into account are both quantum annealing and circuit-based universal quantum computing. The results obtained have shown the effectiveness of most of the proposed approaches.