Dorta, Garoe, Yang, Yongliang, and Campbell, Neill
machine learning, deep neural networks, image editing, uncertainty, and probability
This thesis addresses the task of photo-realistic semantic image editing, where the goal is to provide intuitive controls to modify the content of an image, such that the result is indistinguishable from a real image. In particular the focus is on editing applied to human faces, although, the proposed models can be readily applied to other type of images. We build on recently proposed deep generative models, which allow learning the image editing operations from data. However, there are a number of limitations in these models, two of which are explored in this thesis: the difficulty of modelling high-frequency image details, and the inability to edit images at arbitrarily high resolutions. The difficulty of modelling high-frequency image details is typical of methods with explicit likelihoods. This work presents a novel approach to overcome this problem. This is achieved by surpassing the common assumption that the pixels in the image noise distribution are independent. In most scenarios, breaking away from this independence assumption leads to a significant increase in computational costs. Additionally, it introduces issues in the estimability of the distribution due to the considerable increment in the number of parameters to be estimated. To overcome these obstacles, we present a tractable approach for a correlated multivariate Gaussian data likelihood, based on sparse inverse covariance matrices. This approach is demonstrated on variational autoencoder (VAE) networks. An approach to perform image edits using generative adversarial networks (GAN) at arbitrarily high-resolutions is also proposed. The method relies on restricting the types of edits to smooth warps, i.e. geometric deformations of the input image. These warps can be efficiently learned and predicted at a lower resolution, and easily upsampled to be applied at arbitrary resolutions with minimal loss of fidelity. Moreover, paired data is not needed for training the method, i.e. example images of the same subject with different semantic attributes. The model offers several advantages with respect to previous approaches that directly predict the pixel values: the edits are more interpretable, the image content is better preserved, and partial edits can be easily applied.
Clayton, Robert Vivian, Borthwick, Alistair, Ingram, David, and Smith, H.
tidal energy, renewable, uncertainty, hydrodynamic modelling, harmonic analysis, and ADCP
Tidal stream energy has the potential to contribute to a diverse future energy mix. As the industry moves towards commercialisation and array scale deployment, there is an opportunity to better understand the uncertainties around energy yield assessments. Energy yield assessments are used widely in the wind industry to evaluate the potential energy production from a prospective project. One of the key challenges is to quantify and reduce uncertainty in energy yield assessment. This thesis investigates ways to achieve this through utilising lessons learnt from the established wind industry. An evaluation of both the wind and tidal energy yield assessment process is conducted, highlighting where synergies can be used to increase understanding of uncertainty for the nascent tidal industry. The processes are comparable starting with a campaign to collect site data to characterise the resource at the measurement location. The next stage is to evaluate the long term variations, however this is where the two methods differ. Analysis of long term wind effects requires correlations to be made between short term site data and long term reference data from alternative sources. An assessment of tidal variations over longer periods utilises harmonic analysis, which is capable of deconstructing the individual astronomical variations of the tide and reconstructing them to predict future variations. Despite harmonic analysis being able to determine the astronomical effects of the tide, there are uncertainties in the measurements of tidal flow which are associated with non-astronomical effects. Effects such as turbulence introduce uncertainty when evaluating measured tidal data. This is one area which is investigated further in the thesis. Methods to evaluate the turbulence intensity from real ADCP data are investigated. The next stages require creating a numerical model of the site to extrapolate the data spatially to other areas of interest (such as a turbine location). Energy yield predictions for both wind and tidal are made by combining a power curve with the long term resource. The energy yield outputs are then adjusted to account for energy losses and uncertainties are applied to produce final energy yield values with the attributed probability values associated. Statistical methods are applied to harmonic analysis to assess the level of uncertainty in long term predictions of tidal variations. A method using spectral analysis is applied to evaluate the residuals between measured and modelled data and proves to be accurate at determining missing tidal constituents from the analysis. A method for evaluating the turbulence intensity of the flow is shown, to better understand the stochastic nature of the tidal signal. An investigation is conducted to assess the propagation of bed friction uncertainty, in hydrodynamic modelling, and the resulting impact on the predicted power output from a theoretical fence of tidal turbines spanning a tidal channel. The methodology is based on first conducting sensitivity studies by varying a parameter in the model and calculating the power. Then using a mean and standard deviation for the input parameter, the impact of the uncertainty can be transferred to the estimate of power. The results show that a larger uncertainty associated with the bed roughness tends to over predict the estimation of power. This work aims to inform the standardisation of practices and guidelines in tidal resource assessment and to support developers, consultants and financiers in future tidal energy yield assessments. The final chapter includes procedural recommendations for future tidal energy projects, summarising methods to calculate uncertainty and recommendations to reduce them.