IT Licentiate theses
http://www.it.uu.se/research/publications/lic
Licentiate theses from the Department of Information Technology, Uppsala University, Sweden
20240520T11:37:56Z
Department of Information Technology, Uppsala University, Sweden
Copyright © 2005 Department of Information Technology, Uppsala University, Sweden
daily
12
19990101T00:00+01:00

Licentiate thesis 2024001: Capital and the Social Reproduction of Inequality in Computing Education
http://www.it.uu.se/research/publications/lic/2024001
20240531
Thom Kunkeler
<b>Abstract:</b> Computing education in Western countries has traditionally been characterised by low levels of participation and diversity among its student population. In order to broaden participation in the field, it is fundamental to understand the various mechanisms through which power structures and inequality are reproduced. From a Bourdieusian perspective, this licentiate thesis sets out to understand the interaction between capital, class, and habitus which allows a dominant class to thrive at the expense of other classes. Paper I shows that capital serves as a barrier for noncomputing students entering the computing field, whereas in Paper II a dominant class is identified as possessing higher levels of capital, which is then related to their higher levels of participation in the field. In addition, Paper I provides insight into the ways the subordinate class internalises and acts upon their lower levels of capital. This licentiate thesis lays out the groundwork for studying capital in computing education by developing and validating research instruments which can be used for further study. In addition, relevant theories to educational participation are discussed, with a particular focus on capital theory. More work is needed to understand the reproductive mechanism through which the dominant class legitimises their capital within the field of computing education, thereby establishing their class position. Future work is recommended in the domain of habitus and capitalinclusive pedagogy. Ultimately, the goal is to reduce the reproduction of inequality in computing education by assessing the various mechanisms involved, and designing pedagogy which can be used for successful engagement of students with varying levels of capital.

Licentiate thesis 2023003: Security Allocation in Networked Control Systems
http://www.it.uu.se/research/publications/lic/2023003
20231013
Anh Tung Nguyen
<b>Abstract:</b> Sustained use of critical infrastructure, such as electrical power and water distribution networks, requires efficient management and control. Facilitated by the advancements in computational devices and nonproprietary communication technology, such as the Internet, the efficient operation of critical infrastructure relies on network decomposition into interconnected subsystems, thus forming networked control systems. However, the use of public and pervasive communication channels leaves these systems vulnerable to cyber attacks. Consequently, the critical infrastructure is put at risk of suffering operation disruption and even physical damage that would inflict financial costs as well as pose a hazard to human health. Therefore, security is crucial to the sustained efficient operation of critical infrastructure. This thesis develops a framework for evaluating and improving the security of networked control systems in the face of cyber attacks. The considered security problem involves two strategic agents, namely a malicious adversary and a defender, pursuing their specific and conflicting goals. The defender aims to efficiently allocate defense resources with the purpose of detecting malicious activities. Meanwhile, the malicious adversary simultaneously conducts cyber attacks and remains stealthy to the defender. We tackle the security problem by proposing a gametheoretic framework and characterizing its main components: the payoff function, the action space, and the available information for each agent. Especially, the payoff function is characterized based on the outputtooutput gain security metric that fully explores the worstcase attack impact. Then, we investigate the properties of the game and how to efficiently compute its equilibrium. Given the combinatorial nature of the defender's actions, one important challenge is to alleviate the computational burden. To overcome this challenge, the thesis contributes several systemand graphtheoretic conditions that enable the defender to shrink the action space, efficiently allocating the defense resources. The effectiveness of the proposed framework is validated through numerical examples.

Licentiate thesis 2023002: Integrating Prior Knowledge into Machine Learning Models with Applications in Physics
http://www.it.uu.se/research/publications/lic/2023002
20230920
Philipp Pilar
<b>Abstract:</b> At the extremes, two antithetical approaches to describing natural processes exist. Theoretical models can be derived from first principles, allowing for clear interpretability; on the downside, this approach may be infeasible or inefficient for complex systems. Alternatively, methods from statistical machine learning can be employed to learn black box models from large amounts of data, while providing little or no understanding of their inner workings. Both approaches have different desirable properties and weaknesses. It is natural to ask how they may be combined to create better models. This is the question that the field of physicsinformed machine learning is concerned with, and which we will consider in this thesis. More precisely, we investigate ways of integrating additional prior knowledge into machine learning models. In Paper I, we consider multitask Gaussian processes and devise a way to include socalled sum constraints into the model, where a nonlinear sum of the outputs is required to equal a known value. In Paper II, we consider the task of determining unknown parameters from data when solving partial differential equations (PDEs) with physicsinformed neural networks. Given the prior knowledge that the measurement noise is homogeneous but otherwise unknown, we demonstrate that it is possible to learn the solution and parameters of the PDE jointly with the noise distribution. In Paper III, we consider generative adversarial networks, which may produce realisticlooking samples but fail to reproduce their true distribution. In our work, we mitigate this issue by matching the true and generated distributions of statistics extracted from the data.

Licentiate thesis 2023001: Modeling and Estimation of Impulsive Biomedical Systems
http://www.it.uu.se/research/publications/lic/2023001
20230612
Håkan Runvik
<b>Abstract:</b> Dynamical systems are often expressed in either continuous or discrete time. Some biomedical processes are however more suitably modeled as impulsive systems, which combine continuous dynamics with abrupt changes of the state of the system. This thesis concerns two such systems: the pharmacokinetics of the antiParkinson's drug levodopa, and the testosterone regulation in the human male. Despite the differences between these systems, they can be modeled in similar ways. Modeling entails not only the model, but also the methods used to estimate its parameters. Impulsive dynamics can enable simpler representations compared with using continuous dynamics alone, but may also complicate the estimation procedure, since standard techniques often cannot be used. The contributions of this thesis are therefore both in model development and parameter estimation. Model development is the topic of Paper I. It presents a model of the multipeaking phenomenon in levodopa pharmacokinetics, which is manifested by secondary concentration peaks in the blood concentration profile of the drug. The remaining papers focus on estimation, in a setup where a sequence of impulses is fed to a linear plant, whose output is measured. Two estimation techniques are considered. The first is presented in Paper II and uses a Laguerre domain representation to estimate the timing and weights of the impulses. The second combines estimation of the impulsive input with estimation of the plant parameters, which represent the elimination rates of testosteroneregulating hormones. This problem is particularly challenging since increasing the estimated elimination rates and the number of impulses generally improves the model fit, but only models with sparse input signals are practically useful. Paper III addresses this issue through a novel regularization method. The uncertainties in model and measurements encountered when working with clinical hormone data add another layer of complexity to the problem; methods for handling such issues are described in Paper IV.

Licentiate thesis 2022003: Computational Statistical Methods for Genotyping Biallelic DNA Markers from Pooled Experiments
http://www.it.uu.se/research/publications/lic/2022003
20221109
Camille Clouard
<b>Abstract:</b> The information conveyed by genetic markers such as Single Nucleotide Polymorphisms (SNPs) has been widely used in biomedical research for studying human diseases, but also increasingly in agriculture by plant and animal breeders for selection purposes. Specific identified markers can act as a genetic signature that is correlated to certain characteristics in a living organism, e.g. a sensitivity to a disease or highyield traits. Capturing these signatures with sufficient statistical power often requires large volumes of data, with thousands of samples to analyze and possibly millions of genetic markers to screen. Establishing statistical significance for effects from genetic variations is especially delicate when they occur at low frequencies. The production cost of such marker genotype data is therefore a critical part of the analysis. Despite recent technological advances, the production cost can still be prohibitive and genotype imputation strategies have been developed for addressing this issue. The genotype imputation methods have been widely investigated on human data and to a smaller extent on crop and animal species. In the case where only few reference genomes are available for imputation purposes, such as for nonmodel organisms, the imputation results can be less accurate. Group testing strategies, also called pooling strategies, can be wellsuited for complementing imputation in large populations and decreasing the number of genotyping tests required compared to the single testing of every individual. Pooling is especially efficient for genotyping the lowfrequency variants. However, because of the particular nature of genotype data and because of the limitations inherent to the genotype testing techniques, decoding pooled genotypes into unique data resolutions is a challenge. Overall, the decoding problem with pooled genotypes can be described as as an inference problem in Missing Not At Random data with nonmonotone missingness patterns. Specific inference methods such as variations of the ExpectationMaximization algorithm can be used for resolving the pooled data into estimates of the genotype probabilities for every individual. However, the nonrandomness of the undecoded data impacts the outcomes of the inference process. This impact is propagated to imputation if the inferred genotype probabilities are to be devised as input into classical imputation methods for genotypes. In this work, we propose a study of the specific characteristics of a pooling scheme on genotype data, as well as how it affects the results of imputation methods such as treebased haplotype clustering or coalescent models.

Licentiate thesis 2022002: Making Sampled Simulations Faster by Minimizing Warming Time
http://www.it.uu.se/research/publications/lic/2022002
20221028
Gustaf Borgström
<b>Abstract:</b> A computer system simulator is a fundamental tool for computer architects to try out brand new ideas or explore the system's response to different configurations when executing different program codes. However, even simulating the CPU core in detail is timeconsuming as the execution rate slows down by several orders of magnitude compared to native execution. To solve this problem, previous work, namely SMARTS, demonstrates a statistical sampling methodology that records measurements only from tiny samples throughout the simulation. It spends only a fraction of the full simulation time on these sample measurements. Inbetween detailed sample simulations, SMARTS fastforwards in the simulation using a greatly simplified and much faster simulation model (compared to full detail), which maintains only necessary parts of the architecture, such as cache memory. This maintenance process is called warming. While warming is mandatory to keep the simulation accuracy high, caches may be sufficiently warm for an accurate simulation long before reaching the sample. In other words, much time may be wasted on warming in SMARTS. In this work, we show that caches can be kept in an accurate state with much less time spent on warming. The first paper presents Adaptive Cache Warming, a methodology for identifying the minimum amount of warming in an iterative process for every SMARTS sample. The rest of the simulation time, previously spent on warming, can be skipped by fastforwarding between samples using native hardware execution of the code. Doing so will thus result in significantly faster statistically sampled simulation while maintaining accuracy. The second paper presents Cache Merging, which mitigates the redundant warmings introduced in Adaptive Cache Warming. We solve this issue by going back in time and merging the existing warming with a cache warming session that comes chronologically before the existing warming. By removing the redundant warming, we yield even more speedup. Together, Adaptive Cache Warming and Cache Merging is a powerful boost for statistically sampled simulations.

Licentiate thesis 2022001: Secure Inbody Communication and Sensing
http://www.it.uu.se/research/publications/lic/2022001
20221026
Sam Hylamia
<b>Abstract:</b> Implantable medical devices (IMDs) such as cardiac implants and insulin pumps provide patients with lifesaving functions and improve their lives. These properties make them an integral part of medical professionals' toolbox. Today, IMDs which can be controlled or adjusted wirelessly are widely adopted and are becoming increasingly connected to each other and to the internet. While the modern communication properties of IMDs provide substantial benefits, they pose a major cybersecurity risk when devices are not secured adequately. In this thesis, we explore security issues related to the communication and sensing capabilities of modern onbody devices such as IMDs. In particular, we investigate authentication and key agreement in a network of bodyworn devices, and address the privacy of inbody continuous sensing and monitoring. The main contributions of this thesis are twofold: (1) We propose and evaluate Tiek, an authentication and key distribution protocol for networked bodyworn devices. Tiek authenticates the presence of participating devices on the body and distributes cryptographic keys to them using environment based sources of randomness. The protocol utilizes a twotier authorization scheme to restrict the access of malbehaving bodyworn participants to the network. (2) We also study the information leakage associated with the deployment of a novel inbody continuous monitoring technique. We target the information leakage from the sensing process, and propose and evaluate privacy enhancing measures that prevent a passive eavesdropper from violating the privacy of the patient. We believe this thesis contributes to the development of secure onbody devices in general and IMDs in particular.

Licentiate thesis 2021002: Improving Training of Deep Learning for Biomedical Image Analysis and Computational Physics
http://www.it.uu.se/research/publications/lic/2021002
20211222
Karl Bengtsson Bernander
<b>Abstract:</b> The previous decade has seen breakthroughs in image analysis and computer vision, mainly due to machine learning methods known as deep learning. These methods have since spread to other fields. This thesis aims to survey the progress, highlight problems related to data and computations, and show techniques to mitigate them. In Paper I, we show how to modify the VGG16 classifier archi tecture to be equivariant to transformations in the p4 group, consisting of translations and specific rotations. We conduct experiments to investigate if baseline architectures, using data augmentation, can be replaced with these rotationequivariant networks. We train and test on the Oral cancer dataset, used to automate cancer diagnostics. In Paper III, we use a similar methodology as in Paper I to modify the Unet architecture combined with a discriminative loss, for semantic instance segmentation. We test the method on the BBBC038 dataset consisting of highly varied images of cell nuclei. In Paper II, we look at the UCluster method, used to group sub atomic particles in particle physics. We show how to distribute the training over multiple GPUs using distributed deep learning in a cloud environment. The papers show how to use limited training data more effi ciently, using groupequivariant convolutions, to reduce the prob lems of overfitting. They also demonstrate how to distribute training over multiple nodes in computational centers, which is needed to handle growing data sizes.

Licentiate thesis 2021001: On the Registration and Modeling of Sequential Medical Images
http://www.it.uu.se/research/publications/lic/2021001
20211216
Niklas Gunnarsson
<b>Abstract:</b> Realtime imaging can be used to monitor, analyze and control medical treatments. In this thesis, we want to explain the spatiotemporal motion and thus enable more advanced procedures, especially realtime adaptation in radiation therapy. The motion occurring between image acquisitions can be quantified by image registration, which generates a mapping between the images. The contribution of the thesis consists of three papers, where we have used different approaches to estimate the motion between images. In Paper I, we combine a stateoftheart method in realtime tracking with a learned sparsetodense interpolation scheme. For this, we track an arbitrary number of regions in a sequence of medical images. We estimated a sparse displacement field, based on the tracking positions and used the interpolation network to achieve its dense representation. Paper II was a contribution to a challenge in learnable image registration where we finished at 2nd place. Here we train a deep learning method to estimate the dense displacement field between two images. For this, we used a network architecture inspired by both conventional medical image registration methods and optical flow in computer vision. For Paper III, we estimate the dynamics of spatiotemporal images by training a generative network. We use nonlinear dimensional reduction techniques and assume a linear dynamic in a lowdimensional latent space. In comparison with conventional image registration methods, we provide a method more suitable for realworld scenarios, with the possibility of imputation and extrapolation. Although the problem is challenging and several questions are left unanswered we believe a combination of conventional, learnable, and dynamic modeling of the motion is the way forward.

Licentiate thesis 2020006: Calibration of Probabilistic Predictive Models
http://www.it.uu.se/research/publications/lic/2020006
20201028
David Widmann
<b>Abstract:</b> Predicting unknown and unobserved events is a common task in many domains. Mathematically, the uncertainties arising in such prediction tasks can be described by probabilistic predictive models. Ideally, the model estimates of these uncertainties allow us to distinguish between uncertain and trustworthy predictions. This distinction is particularly important in safetycritical applications such as medical image analysis and autonomous driving. For the probabilistic predictions to be meaningful and to allow this differentiation, they should neither be over nor underconfident. Models that satisfy this property are called calibrated. In this thesis we study how one can measure, estimate, and statistically reason about the calibration of probabilistic predictive models. In Paper I we discuss existing approaches for evaluating calibration in multiclass classification. We mention potential pitfalls and suggest hypothesis tests for the statistical analysis of model calibration. In Paper II we propose a framework of calibration measures for multiclass classification. It captures common existing measures and includes a new kernel calibration error based on matrixvalued kernels. For the kernel calibration error consistent and unbiased estimators exist and asymptotic hypothesis tests for calibration can be derived. Unfortunately, by construction the framework is limited to prediction problems with finite discrete target spaces. In Paper III we use a different approach to develop a more general framework of calibration errors that applies to any probabilistic predictive model and is not limited to classification. We show that it coincides with the framework presented in Paper II for multiclass classification. Based on scalarvalued kernels, we generalize the kernel calibration error, its estimators, and hypothesis tests to all probabilistic predictive models. For realvalued regression problems we present empirical results.

Licentiate thesis 2020005: Exploiting Conjugacy in StateSpace Models with Sequential Monte Carlo
http://www.it.uu.se/research/publications/lic/2020005
20200514
Anna Wigren
<b>Abstract:</b> Many processes we encounter in our daily lives are dynamical systems that can be described mathematically using statespace models. Exact inference of both states and parameters in these models is, in general, intractable. Instead, approximate methods, such as sequential Monte Carlo and Markov chain Monte Carlo, are used to infer quantities of interest. However, samplebased inference inherently introduces variance in the estimates. In this thesis we explore different aspects of how conjugacy relations in a model can improve the performance of sequential Monte Carlobased inference methods. A conjugacy relation between the prior distribution and the likelihood implies that the posterior distribution has the same distributional form as the prior, allowing for analytic updates in place of numerical integration. In Paper I we consider state inference in statespace models where the transition density is intractable. By adding artificial noise conjugate to the observation density we can design an efficient proposal for sequential Monte Carlo inference that can reduce the variance of the state estimates. Conjugacy can also be utilized in the setting of parameter inference. In Paper II we show that the performance of particle Gibbstype samplers, in terms of the autocorrelation of the samples, can be improved when conjugacy relations allow for marginalizing out the dependence on parameters in the state update. Despite enabling analytical evaluation of integrals, the derivation and implementation of conjugacy updates is cumbersome in all but the simplest cases, which limits the usefulness in practice. Recently, the emerging field of probabilistic programming has changed this, by providing a framework for automating inference in probabilistic models  including identifying and utilizing conjugacy relations. In Paper II we make use of probabilistic programming to automatically exploit conjugacy in an epidemiological statespace model describing the spread of dengue fever.

Licentiate thesis 2020004: Machine Learning for Spatially Varying Data
http://www.it.uu.se/research/publications/lic/2020004
20200422
Muhammad Osama
<b>Abstract:</b> Many physical quantities around us vary across space or spacetime. An example of a spatial quantity is provided by the temperature across Sweden on a given day and as an example of a spatiotemporal quantity we observe the counts of the corona virus cases across the globe. Spatial and spatiotemporal data enable opportunities to answer many important questions. For example, what the weather would be like tomorrow or where the highest risk for occurrence of a disease is in the next few days? Answering questions such as these requires formulating and learning statistical models. One of the challenges with spatial and spatiotemporal data is that the size of data can be extremely large which makes learning a model computationally costly. There are several means of overcoming this problem by means of matrix manipulations and approximations. In paper I, we propose a solution to this problem where the model is learned in a streaming fashion i.e. as the data arrives point by point. This also allows for efficient updating of the learned model based on newly arriving data which is very pertinent to spatiotemporal data. Another interesting problem in the spatial context is to study the causal effect that an exposure variable has on a response variable. For instance, policy makers might be interested in knowing whether increasing the number of police in a district has the desired effect of reducing crimes there. The challenge here is that of spatial confounding. A spatial map of the number of police against the spatial map of the number of crimes in different districts might show a clear association between these two quantities. However, there might be a third unobserved confounding variable that makes both quantities small and large together. In paper II, we propose a solution for estimating causal effects in the presence of such a confounding variable. Another common type of spatial data is point or event data, i.e., the occurrence of events across space. The event could for example be a reported disease or crime and one may be interested in predicting the counts of the event in a given region. A fundamental challenge here is to quantify the uncertainty in the predicted counts in a model in a robust manner. In paper III, we propose a regularized criterion for learning a predictive model of counts of events across spatial regions. The regularization ensures tighter prediction intervals around the predicted counts and have valid coverage irrespective of the degree of model misspecification.

Licentiate thesis 2020003: Securing the Memory Hierarchy from Speculative SideChannel Attacks
http://www.it.uu.se/research/publications/lic/2020003
20200306
Christos Sakalis
<b>Abstract:</b> Modern highperformance CPUs depend on speculative outoforder execution in order to offer high performance while also remaining energy efficient. However, with the introduction of Meltdown and Spectre in the beginning of 2018, speculative execution has been under attack. These attacks, and the many that followed, take advantage of the unchecked nature of speculative execution and the microarchitectural changes it causes in order to mount speculative sidechannel attacks. Such attacks can bypass software and hardware barriers and gain access to sensitive information while remaining invisible to the application. In this thesis we will describe our work on preventing speculative sidechannel attacks that exploit the memory hierarchy as their sidechannel. Specifically, we will discuss two different approaches, one that does not restrict speculative execution but tries to keep its microarchitectural sideeffects hidden, and one where we delay speculative memory accesses if we determine that they might lead to information leakage. We will discuss the advantages and disadvantages of both approaches, compare them against other stateoftheart solutions, and show that it is possible to achieve secure, invisible speculation while at the same time maintaining high performance and efficiency.

Licentiate thesis 2020002: Global Radial Basis Function Collocation Methods for PDEs
http://www.it.uu.se/research/publications/lic/2020002
20200320
Ulrika Sundin
<b>Abstract:</b> Radial basis function (RBF) methods are meshfree, i.e., they can operate on unstructured node sets. Because the only geometric information required is the pairwise distance between the node points, these methods are highly flexible with respect to the geometry of the computational domain. The RBF approximant is a linear combination of translates of a radial function, and for PDEs the coefficients are found by applying the PDE operator to the approximant and collocating with the right hand side data. Infinitely smooth RBFs typically result in exponential convergence for smooth data, and they also have a shape parameter that determines how flat or peaked they are, and that can be used for accuracy optimization. In this thesis the focus is on global RBF collocation methods for PDEs, i.e., methods where the approximant is constructed over the whole domain at once, rather than built from several local approximations. A drawback of these methods is that they produce dense matrices that also tend to be illconditioned for the shape parameter range that might otherwise be optimal. One current trend is therefore to use overdetermined systems and least squares approximations as this improves stability and accuracy. Another trend is to use localized RBF methods as these result in sparse matrices while maintaining a high accuracy. Global RBF collocation methods together with RBF interpolation methods, however, form the foundation for these other versions of RBFPDE methods. Hence, understanding the behaviour and practical aspects of global collocation is still important. In this thesis an overview of global RBF collocation methods is presented, focusing on different versions of global collocation as well as on method properties such as error and convergence behaviour, approximation behaviour in the small shape parameter range, and practical aspects including how to distribute the nodes and choose the shape parameter value. Our own research illustrates these different aspects of global RBF collocation when applied to the Helmholtz equation and the BlackScholes equation.