| 11. |
- Andersson, Carl
(författare)
-
Deep probabilistic models for sequential and hierarchical data
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Consider the problem where we want a computer program capable of recognizing a pedestrian on the road. This could be employed in a car to automatically apply the brakes to avoid an accident. Writing such a program is immensely difficult but what if we could instead use examples and let the program learn what characterizes a pedestrian from the examples. Machine learning can be described as the process of teaching a model (computer program) to predict something (the presence of a pedestrian) with help of data (examples) instead of through explicit programming.This thesis focuses on a specific method in machine learning, called deep learning. This method can arguably be seen as sole responsible for the recent upswing of machine learning in academia as well as in society at large. However, deep learning requires, in human standards, a huge amount of data to perform well which can be a limiting factor. In this thesis we describe different approaches to reduce the amount of data that is needed by encoding some of our prior knowledge about the problem into the model. To this end we focus on sequential and hierarchical data, such as speech and written language.Representing sequential output is in general difficult due to the complexity of the output space. Here, we make use of a probabilistic approach focusing on sequential models in combination with a deep learning structure called the variational autoencoder. This is applied to a range of different problem settings, from system identification to speech modeling.The results come in three parts. The first contribution focus on applications of deep learning to typical system identification problems, the intersection between the two areas and how they can benefit from each other. The second contribution is on hierarchical data where we promote a multiscale variational autoencoder inspired by image modeling. The final contribution is on verification of probabilistic models, in particular how to evaluate the validity of a probabilistic output, also known as calibration.
|
|
| 12. |
- Andersson, Matilda L.
(författare)
-
Fish population responses to climate change : Causes and consequences
- 2021
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Lake environments are heterogeneous, and animals show a variety of adaptations to deal with this heterogeneity. Fish often show intraspecific variation in diet, metabolism, and behavior, corresponding to their habitat use. Studies on climate change often ignore this heterogeneity and its importance in determining population-level responses to climate change. This thesis can be broken into two interacting pieces. First, my goal was to assess how water color and temperature changes impact the size, number, and distribution of a common predator, Eurasian perch (Perca fluviatilis), in Swedish lakes. Second, I aimed to examine whether metabolism and resource use differed between lake habitats, corresponding with documented patterns of polymorphism and whether diet differences were maintained along a thermal and water color gradient. By combining the information gleaned from these studies, the overarching goal of my thesis is to better understand how climate change will impact fish populations and how intraspecific variance in these responses will impact ecosystem functioning. I found that warming and browning will likely decrease fish biomass but via different mechanisms. Warming reduces average fish size through its impact on metabolism and energy requirements. Browning decreases fish abundance likely due to its negative effects on resource abundance, increasing mortality, and decreasing reproductive effort. Though warming decreases biomass at the lake level, pelagic perch abundance increases. I found that these pelagic perch have higher metabolic rates and, especially in darker lakes, rely heavily on pelagic resources. As more fish shift into the pelagic habitat, this will increase top-down pressure on pelagic resources and decrease energy transfer from littoral to pelagic habitats altering energy flow within lakes. Variation in metabolic phenotype across habitats, combined with the positive scaling of metabolic rates with temperature, will likely determine which fish can persist under climate change scenarios. Studies that measure this variation rely heavily on respirometry to measure fish metabolism. I found that current respirometry methods underestimate maximum metabolic rate and suggest an updated method to improve the accuracy of future studies. Overall, I conclude that habitats should be examined separately to better understand population-level responses to climate change. Perch caught in different habitats have different energy requirements and respond differently to warming and browning. These differences will affect the distribution of top-down pressure and habitat coupling within lake ecosystems, with implications for broader ecosystem functioning in the future.
|
|
| 13. |
- Andersson, Rassmus
(författare)
-
Discovering new ground in ion transport: Exploring coordination effects in polymer electrolytes : – From method development to battery implementation
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The exponentially increasing demand for portable and stationary energy storage devices is pushing the development of lithium-ion batteries (LIBs). This requires safer and more sustainable electrolytes where solid polymer electrolytes (SPEs) are a viable alternative to the flammable liquid electrolytes used nowadays. However, SPEs are characterized by poor ionic conductivity compared to their liquid equivalents, preventing large-scale implementation. Furthermore, to meet the increasing production rate of batteries, alternative battery chemistries based on more abundant resources than Li are explored. To address these matters, a fundamental understanding of ion transport in SPEs for a range of relevant cations is vital in the development process.In the thesis, the ion transport is explored on a fundamental level for Li+ in addition to cations “beyond Li” such as Na+, K+ and Mg2+ in polyether-, polyester- and polycarbonate-based SPEs, where the core encompasses the connection between the ion coordination strength and the transference number (T+). New methods to investigate these properties have been developed especially targeting these more challenging cations. To study the ion coordination strength, two qualitative and one quantitative methods based on NMR and FTIR, are presented. In addition, eNMR and EIS have been combined to determine T+.Regardless of the cation investigated, the strongest coordination was observed for polyethylene oxide, stemming from its chelating effect on the cations. In contrast, poly(trimethylene carbonate) exhibited the weakest coordination, while poly(ε-caprolactone) fell in between. A direct correlation between the coordination strength and the T+ was also recognized, where strong interactions are accompanied by low T+ and vice versa. Moreover, the divalent Mg2+ displayed particularly interesting transport characteristics, where the [MgTFSI]+ speciation appears to be a large contributor to the net Mg mobility. Lastly, the outcome of incorporating an ion-conducting polymer as the soft segment in polyurethanes is that the transport mechanism of the pure SPE remains. In combination with sustained long-term cycling in lithium metal batteries, the polyurethanes illustrate opportunities for new designs by adjusting the soft segments. Similarly, the properties of poly(1-oxoheptamethylene) can be controlled by tuning its saturation degree, which is crucial for the ion conduction and mechanical properties in lithium metal batteries, since it highly affects the crystallinity and the crosslinking of the systems.In summary, this thesis contributes toward the understanding of ion transport in systems belonging to “next-generation” batteries, where SPEs for lithium-metal batteries as well as for cations “beyond Li” are considered to play an important part.
|
|
| 14. |
- Andriantsaralaza, Miora, 1996-
(författare)
-
Mass Loss of Evolved Stars : Improving Mass-Loss Rates and Distances
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- In the final stages of their lives, low to intermediate-mass stars enter the Asymptotic Giant Branch (AGB) phase, where they experience significant mass loss through dusty stellar winds. This mass loss is pivotal, not only for the chemical enrichment of the interstellar medium, setting the stage for new stars and planets to form, but also in dictating the evolution and the ultimate fate of the stars themselves. The study of mass loss in AGB stars is therefore of paramount importance. Analyses of emission from CO gas in these outflows allow for the determination of mass-loss rates, as CO serves as a primary tracer for the molecular gas in the circumstellar envelopes of AGB stars. However, one of the main uncertainties in mass loss estimations arises from the assumptions on the size of the CO envelope. This uncertainty can be removed using interferometry, which allows for direct measurements of the spatial extent of the CO emission. Using the compact array of the Atacama Large Millimeter/submillimeter Array (ALMA), we measured the extent and investigated the degree of sphericity of the CO-emitting regions around 69 AGB stars in the DEATHSTAR programme. Of those, 27 are presented in this thesis.Another significant source of uncertainty in mass-loss rate estimates lies in the distance, a fundamental parameter which has been notoriously difficult to accurately determine for AGB stars, especially when relying on optical parallaxes like those from Gaia. To tackle this, we conducted a comparative analysis between Gaia DR3 parallaxes and the more robust parallaxes obtained from high-resolution interferometric observations of maser emissions. This approach enabled us to provide reliable distance estimates for a sample of 200 AGB stars, including the DEATHSTAR sample, offering a valuable resource for the AGB scientific community. Using the newly calculated distances and updated CO envelope size measurements, we performed radiative transfer modelling to derive the mass-loss rates of a sample of 27 carbon-rich AGB stars. The spatial information provided by ALMA observations acted as strong constraints for our models, ensuring a more accurate representation of flux contributions from various spatial scales.By providing the measurement of CO envelope sizes, improving the distance determination to AGB stars, and presenting improved mass-loss rates, this thesis provides new insights into the complex nature of AGB stars and their mass loss.
|
|
| 15. |
- Artemov, Anton G., 1990-
(författare)
-
Parallelization of dynamic algorithms for electronic structure calculations
- 2021
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The aim of electronic structure calculations is to simulate behavior of complex materials by resolving interactions between electrons and nuclei in atoms at the level of quantum mechanics. Progress in the field allows to reduce the computational complexity of the solution methods to linear so that the computational time scales proportionally to the size of the physical system. To solve large scale problems one uses parallel computers and scalable codes. Often the scalability is limited by the data distribution.This thesis focuses on a number of problems arising in electronic structure calculations, such as inverse factorization of Hermitian positive definite matrices, approximate sparse matrix multiplication, and density matrix purification methods. No assumptions are made about the data distribution, instead, it is explored dynamically.The thesis consists of an introduction and five papers. Particularly, in Paper I we present a new theoretical framework for localized matrices with exponential decay of elements. We describe a new localized method for inverse factorization of Hermitian positive definite matrices. We show that it has reduced communication costs compared to other widely used parallel methods. In Paper II we present a parallel implementation of the method within the Chunks and Tasks programming model and do a scalability analysis based on critical path length estimation.We focus on the density matrix purification technique and its core operation, sparse matrix-matrix multiplication, in Papers III and IV. We analyze the sparse approximate matrix multiplication algorithm with the proposed localization framework, add a prior truncation step, and derive the asymptotic behavior of the Frobenius norm of the error. We employ the sparse approximate multiplication algorithm in the density matrix purification process and propose a method to control the error norm by choosing the right truncation threshold value. We present a new version of the Chunks and Tasks matrix library in Paper V. The library functionality and architecture are described and discussed. The efficiency of the library is demonstrated in a few computational experiments.
|
|
| 16. |
- Asmuth, Henrik
(författare)
-
Efficient Large-eddy Simulation for Wind Energy Applications
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Modelling the interaction of wind turbines with the ambient flow is essential for almost all technical aspects of wind energy exploitation. Large-eddy simulation (LES) is the most detailed approach feasible to model this complex interaction of wind turbines with the atmospheric boundary layer and the wakes of upstream turbines. Despite more than twenty years of fundamental research on wind turbine modelling with LES, applications of the method remain limited to academic use cases to date. The main bottleneck hindering a broader adoption of LES in the industrial practice is the large computational demand of the method. Nevertheless, it holds enormous potential for addressing various modelling challenges arising from current trends in wind energy.A promising alternative to classical numerical approaches for LES is the lattice Boltzmann method (LBM). In particular, GPU-based (graphics processing unit) implementations of the method provide significant performance gains and have enabled unprecedented computational efficiencies for LES in different fields of fluid dynamics. Still, the LBM´s potential for wind energy applications remains untapped due to open questions, some of which are specific to the field. This thesis addresses two specific problems in applications of LES to wind turbine and farm simulations. First, is the representation of wind turbines with the actuator line technique. And, second, is the modelling of the surface shear stress in simulations of atmospheric boundary layers. Both aspects are crucial to enable LES for wind energy applications with the LBM, as is usually done with conventional approaches.As for the former, an LBM implementation of the actuator line model is applied in multiple studies on wind turbine wakes. Code-to-code comparisons and experimental validations show that the model can accurately capture the aerodynamic forces acting on the turbine blades as well as the wake characteristics. For the simulation of boundary layer flows a novel LBM-specific wall model is developed. The model, referred to as inverse momentum exchange method, imposes the surface shear stress at the first offwall grid points by adjusting the slip velocity in bounce-back boundary schemes. Simulations are compared to theoretical, numerical, and experimental reference data of isothermal boundary layer flows. It is consistently found that both mean quantities and higherorder turbulence statistics can be well-captured by wall-modelled lattice Boltzmann LES using the presented wall model and the employed cumulant collision scheme.The results presented illustrate that the LBM is a suitable approach for state-of-the-art LES of wind turbine wakes and boundary layer flows. Moreover, the applied method is shown to be robust, and, above all, extremely computationally efficient. Based on the observed computational efficiencies, it is concluded that industry LES for wind energy applications is possible with GPU-based LBM solvers. Furthermore, additional studies presented in this thesis illustrate further potentials of the method. Such are applications of reinforcement learning to wind farm control or large-scale data generation for the training of deep learning models for wake predictions.
|
|
| 17. |
- Augstein, Frauke
(författare)
-
Mechanisms of plant root xylem developmental plasticity in response to water deficiency and salt
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Plants may be exposed to a variety of different environmental conditions including water deficiency and salt, both affecting the uptake of water into the plant. Water is taken up from the soil by the roots and distributed throughout the plant via the water conducting tissue, the xylem. Plants are remarkably plastic and have evolved different mechanisms to sense the environment and adjust their development accordingly. However, how xylem development may respond to water availability is not clear. In this thesis, I show how water deficiency and salt affect xylem development and how the observed phenotypic alterations are regulated on a molecular level. We found that upon water deficiency additional protoxylem strands were formed along with an early differentiation of the inner metaxylem. These phenotypes were regulated both by non-cell autonomous and cell autonomous signaling via the hormone abscisic acid (ABA). The expression of microRNA165 was induced by ABA signaling in the endodermis leading to downregulation of homeo domain leucine zipper class III (HD-ZIP III) transcription factors in the stele. This caused a shift in xylem identity from meta- to protoxylem and the formation of additional protoxylem strands. At the same time, cell autonomous ABA signaling upregulated several VASCULAR RELATED NAC DOMAIN (VND) transcription factors including VND7, which promoted the shift in xylem identity as well as VND2 and VND3, which promoted early differentiation of the inner metaxylem. In contrast, during an initial phase of salt stress, we observed the formation of protoxylem gaps specifically in response to ionic stress and distinct from ABA-signaling. We identified that protoxylem gaps were caused by lowered levels and signaling of the growth regulator gibberellin (GA). Downstream of GA-signaling, protoxylem gap formation upon salt was controlled by genes involved in secondary cell wall formation including the xylem master regulator VND6 and factors involved in cell wall modification. Salt tolerance assays suggested that protoxylem gaps may contribute to salt tolerance and the phenotypes that we observed upon water deficiency have been suggested to confer drought tolerance. We observed similar effects on xylem developmental plasticity in response to water deficiency and salt in various different dicot species indicating an evolutionary conservation. Thus, xylem development is of high relevance for breeding programs to generate plant varieties better adapted to a changing climate.
|
|
| 18. |
- Ayala, Ana I.
(författare)
-
Modelling impact climate-related change on the thermal responses of lakes
- 2023
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- In response to climate-related changes, lakes worldwide have experienced warmer surface water temperatures, shorter ice cover periods and changes in lake stratification. As these aspects of lake dynamics exert substantial control over nutrient availability, oxygenation and biogeochemical cycling, predicting changes in lake water temperature and stratification dynamics can improve our understanding of the consequences of warming on lake ecosystems. This thesis investigates the long-term and short-term (extreme event) effects of climate change on lake thermal dynamics using 1D hydrodynamic lake models.Long-term lake water temperature simulations showed that water temperatures and thermal stratification metrics were projected to clearly shift toward lake thermal conditions that are consistent with a warmer climate at the end of the 21st century, i.e. warmer surface and bottom temperatures and a stronger and longer duration of summer thermal stratification as a result of an earlier onset of stratification and later fall overturn. The simulated lake thermal structure was controlled by energy exchange between the lake surface and the atmosphere (surface heat fluxes) and wind stress. The individual surface heat flux components were projected to change substantially under future climate scenarios. However, the combined changes showed compensating effects, leading to a small overall change in total surface heat flux, that was still sufficient to lead to important changes in whole-lake temperature. On a seasonal scale, spring heating and autumnal cooling were projected to decrease, while only small changes were projected in winter and summer. An extended analysis during summer using 47 lakes showed that while all lakes gained heat during summer under all scenarios, differences in the amount of heat gained during historical and future conditions were small. Additionally, hydrodynamic lake models performed well in reproducing the magnitude and direction of changes in lake temperature and stratification metrics during storms and heatwaves. However, the lake model performance decreased in accuracy compared to non-extreme condition, which should be taken into account. 1D hydrodynamic lake models have been shown to be powerful tools to predict long-term and short-term climate-related changes in lake thermal dynamics, making an in-depth analysis of the surface heat fluxes possible.
|
|
| 19. |
- Aziz, Imran
(författare)
-
Ultra-wideband Millimeter-wave Antenna Arrays and Front-end Systems : For high data rate 5G and high energy physics applications
- 2021
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The demand for wireless data communications is rapidly increasing due to several factors including increased internet access, increasingly growing number of mobile users and services, implementation of the Internet of Things (IoT), high-definition (HD) video streaming and video calling. To meet the bandwidth requirement of new and emerging applications, it is necessary to move from the existing microwave bands towards millimeter-wave bands. This thesis presents different antenna arrays at 60 GHz and 28 GHz that are integrated with the front-end RFIC to steer the beam in ≈ ±50° in the azimuth plane. The 5G antenna arrays at 28 GHz are designed to provide broadband high data rate services to the end users. In order to transport this high-volume data to the core network, a fixed wireless access (FWA) link demands the implementation of a broadband, high gain and steerable narrow-beam array. The 60 GHz antenna arrays, presented in this thesis, are good candidates for both FWA as well as backhaul communications. The two proposed arrays at 60 GHz (57-66 GHz) are i) a stacked patches array and ii) a connected slots array feeding a high gain lens antenna. The 2×16 stacked patches antenna array shows more than 20 dBi realized gain. The array is integrated with the front-end RFIC and the resulting module shows > 40 dBm measured effective isotropic radiated power (EIRP). The other 60 GHz antenna array is designed as linear connected slots with sixteen equidistant feeding points. The latest is then used as a feeder of a high gain dielectric lens. Peak measured gain of 25.4 dBi is achieved with this antenna. Moreover, instead of experiencing scan loss, the lens is designed to get higher gain when the beam is steered away from the broadside direction.Furthermore, two compact antenna arrays are designed at 28 GHz (24.25 - 29.50 GHz). A linear polarized (LP) and a circular polarized (CP) array are realized in the fan-out embedded wafer level ball-grid-array (eWLB) package. In comparison with the PCB arrays, this antenna in package (AiP) solution is not only cost-effective but it also reduces the integration losses because of shorter feed lines and no geometrical discontinuity. The LP array is realized as a dipole antenna array feeding a novel horn-shaped heatsink. The RF module gives 34 dBm peak EIRP with beam-steering in ±35°. Besides, the CP antenna array is realized with the help of crossed dipoles and the RF module provides 31 dBm peak EIRP with beam-steering in ±50°.The data demands are not limited to the telecom industry as the upgradation of accelerators and experiments at the large hadron collider (LHC) at CERN will result in increased event rate thus demanding higher data rate front-end readout systems. This work thus investigates the feasibility of 60 GHz wireless links for the data readout at CERN. For this purpose, the 60 GHz wireless chips are irradiated with 17 MeV protons [dose 7.4 Mrad (RX) & 4.2 Mrad (TX)] and 200 MeV electrons [dose 270 Mrad (RX) & 314 Mrad (TX)] in different episodes. The chips have been found operational in the post-irradiation investigations with some performance degradation. The encouraging results motivate to move forward and investigate the realization of wireless links in such a complex environment.
|
|
| 20. |
- Bacic, Luka
(författare)
-
Molecular mechanisms underlying the activation of ALC1 nucleosome remodeling
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Packaging DNA into chromatin represses essential DNA-based processes, such as transcription, DNA replication, and repair. To change the accessibility of DNA, cells have evolved a set of enzymes referred to as chromatin remodelers that act on the basic repeat unit of chromatin, the nucleosome. Chromatin remodelers are critical for normal cell physiology and development. Dysfunction or aberrant regulation of chromatin remodelers can lead to multisystem developmental disorders and cancers. DNA damage represents a major threat to eukaryotic cells. When DNA damage persists, the cell can enter programmed cell death. To avoid such a dramatic outcome, cells must rapidly recognize the DNA damage and trigger DNA repair pathways. An early event following DNA damage is the relaxation of chromatin. Chromatin relaxation depends on ATP consumption and ADP-ribosylation, where the site of DNA damage is marked with ADP-ribose units. ADP-ribose, in turn, can be recognized by the macro domain of the remodeler ALC1 (Amplified in Liver Cancer 1). ALC1 has therefore been implicated in the DNA damage response. In the absence of DNA damage, the macro domain of ALC1 is placed against its ATPase motor to inhibit its activity. However, it is unclear how ALC1, in its active state, engages the nucleosome. Moreover, the mechanism by which ALC1 is fully activated upon recruitment is poorly understood, and the impact of ALC1-catalyzed nucleosome sliding in the vicinity of a DNA damage site is unknown. This thesis investigates how ALC1 engages its substrate, the nucleosome, and how histone modifications can regulate ALC1 activity. Structural and biophysical approaches revealed an ALC1 regulatory segment that binds to the acidic patch, a prominent feature on the nucleosome surface. Further analysis showed that the interaction between ALC1 and the acidic patch is required to fully activate ALC1. Moreover, in vitro ADP-ribosylation of nucleosomes enabled us to form a stable complex of nucleosome-bound ALC1 amenable to structural determination by cryogenic electron microscopy. Our structural models visualize nucleosomal epitopes that play an important role in stimulating productive remodeling by ALC1, as confirmed by various biochemical approaches. Taken together, our data suggested a possible mechanism by which ALC1 could render DNA breaks more accessible to downstream repair factors. Since recent studies defined ALC1 as an attractive anti-cancer target, this thesis provides insights into the molecular mechanisms that regulate ALC1 activity as a potential starting point for structure-based drug development.
|
|
