[diss] Perustieteiden korkeakoulu / SCI

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 20 of 1532
  • Item
    Real-time and sample-efficient learning of computationally rational user models
    (Aalto University, 2024) Keurulainen, Antti; Kaski, Samuel, Prof., Aalto University, Department of Computer Science, Finland; Tietotekniikan laitos; Department of Computer Science; Perustieteiden korkeakoulu; School of Science; Kaski, Samuel, Prof., Aalto University, Department of Computer Science, Finland
    To effectively collaborate with humans, Artificial Intelligence (AI) systems must understand human behavior and the factors influencing it, including their goals, preferences, and abilities. Interactions with humans are typically costly, and in many real-life situations, AI must adapt to human behavior after only a few interactions. Additionally, when AI interacts with humans to learn about their behavior, the interactions need to be conducted without any noticeable delay for the human, which in turn necessitates adaptation in real-time. This thesis investigates how an AI system can learn about other agents in a sample-efficient and real-time manner, using methods based on reinforcement learning. The first contribution of this thesis is a method for learning representations of goal-driven agents' behaviors with neural networks from incomplete observations, showing that they can be used for improving performance in cooperative decision-making tasks. The second contribution concerns the creation of an automated method for producing task distributions and related ground truth data for training a meta-learner to assess the skill level and adapt quickly to the behavior of a cooperating partner. The third contribution presents a novel method for designing informative experiments for estimating the parameters of simulation-based user models without closed-form likelihood functions, and which models are grounded in cognitive science. This method simultaneously amortizes the estimation of these parameters and the designing of experiments. These contributions cover a wide range of settings where useful representations of behavior for improving cooperation are learned, along with the efficient learning of complex user models. The implications of the methods developed, as well as their strengths and limitations, are discussed.
  • Item
    Humans as Information Sources in Bayesian Optimization
    (Aalto University, 2024) Mikkola, Petrus; Tietotekniikan laitos; Department of Computer Science; Aalto Probabilistic Machine Learning Group; Perustieteiden korkeakoulu; School of Science; Kaski, Samuel, Prof., Aalto University, Department of Computer Science, Finland
    Humans are at the heart of the current computational revolution, not only as end-users, but also as integral contributors to computational systems such as machine learning (ML) solutions. This is because these systems depend on data that mainly originate from human activities, such as textual content, artistic creations, or transcribed audio clips. This data is not the only human-derived information flowing into the process, as human expertise plays an important role at all stages of ML development. This thesis reviews methodologies for expert knowledge elicitation, and delves into a promising approach to harnessing humans as a source of information, which is based on the following two ideas. The first idea is to assume the existence of a latent "intuition function" that describes an expert's knowledge over the problem of interest. The intuition function can only be accessed through queries that allow for human feedback, such as preferential queries. Learning the intuition function presents a tractable machine learning problem that can be approached through Gaussian process learning with a probabilistic user model on how the expert data is generated. The second idea pertains to how queries should be selected for an expert and how the expert's knowledge should be applied to the problem of interest. Multi-fidelity Bayesian optimization (MFBO) is a global optimization approach that incorporates multiple information sources with differing levels of accuracy and cost, accelerating the search for optimal solutions. Treating humans as auxiliary information sources within the MFBO framework effectively tackles issues concerning knowledge integration and sample-efficiency. This thesis addresses three problems that arise when humans serve as information sources in Bayesian optimization: (i) the requirement for natural human interaction, (ii) the inherent unreliability of human input, and (iii) the high cost associated with human labor. The articles included in the thesis present novel algorithms as viable solutions to the problems (i), (ii), and (iii). Specifically, we identify problem (ii) as an issue of negative transfer, and we provide an algorithm that establishes theoretical bounds on the negative transfer gap.
  • Item
    Advancing Segmentation of Intracranial Structures in Brain Imaging
    (Aalto University, 2024) Thanellas, Antonios; Ilmoniemi, Risto, Prof., Aalto University, Department of Neuroscience and Biomedical Engineering, Finland; Neurotieteen ja lääketieteellisen tekniikan laitos; Department of Neuroscience and Biomedical Engineering; Perustieteiden korkeakoulu; School of Science; Hämäläinen, Matti, Prof., Aalto University, Department of Neuroscience and Biomedical Engineering, Finland
    There are several challenges and crossroads that a researcher encounters and must navigate while conducting studies in the field of anatomical neuroimaging. These include 1) shortage of large datasets for algorithm testing and the limitations associated with small datasets, 2) finding strategies to make established algorithms more efficient, and 3) development of algorithms robust enough to generalize across different hardware and different acquisition settings. This thesis seeks to address these challenges. It explores the impact of limited datasets on volumetric analyses from Magnetic Resonance Images, introduces innovative approaches for locating intracranial structures and pathologies from Magnetic Resonance (MR) and Computed Tomography (CT) images, and examines how data selection influences analyses and algorithm performance in diverse scenarios. One aspect of the present research quantifies and proposes strategies to mitigate the influence of biases, confounders, and random variations that frequently arise in brain volumetric analyses with limited datasets. The findings emphasise the effectiveness of specific metrics in accurately distinguishing between healthy and non-healthy subjects even in the presence of bias, confounders, or random variation. This thesis also introduces a novel method for segmenting the brain from MR images. This method combines segmentation fusion with a marker-controlled watershed transform and utilizes predictions from established segmentation methods as input. Results demonstrate superior performance compared to the conventional segmentation techniques and other meta-algorithms. In the field of machine learning, the research recommends effective approaches for creating training data with the aim of segmenting intracranial blood from CT images. A neural network developed for this purpose demonstrates potential in segmenting intracranial blood from head CT images while preserving generalizability. In conclusion, the challenges posed by limited datasets require special considerations as they impact the development of both machine learning and classical image processing methods for segmenting structures within the head. While classical image processing methods may see reduced usage on their own, they will likely be increasingly integrated with machine learning approaches.
  • Item
    Spreading and Epidemic Interventions - Effects of Network Structure and Dynamics
    (Aalto University, 2024) K. Rizi, Abbas; Tietotekniikan laitos; Department of Computer Science; Complex Systems Group; Perustieteiden korkeakoulu; School of Science; Kivelä, Mikko, Assist. Prof., Aalto University, Department of Computer Science, Finland
    The COVID-19 pandemic has highlighted the critical importance of understanding epidemic dynamics, particularly the significant gaps in our knowledge that need addressing to better prepare for future pandemics. This thesis delves into the intricacies of disease spread within complex human interaction networks, underlining the pivotal role of individual connectedness in influencing epidemic outcomes. By developing theoretical models inspired by real-world epidemiological data, this work provides a nuanced exploration of disease transmission dynamics across networked populations, emphasizing the heterogeneous, spatial, homophilic, and temporal characteristics inherent in human social structures. A primary focus of this research is the investigation of intervention strategies, encompassing pharmaceutical measures, such as vaccination campaigns, and non-pharmaceutical interventions, including contact tracing techniques. These interventions are evaluated within more realistic network topologies, characterized by degree heterogeneity and group structures, to assess their effectiveness in mitigating epidemic spread. The thesis leverages mathematical and computational epidemiology to offer profound insights into optimizing intervention strategies within the complex web of human interactions, thereby contributing to the academic discourse and providing actionable intelligence for public health policy formulation and epidemic preparedness. The avenues of research opened by this work offer deeper insights into the mechanisms of epidemic spread in social networks. By using stylized modeling, the study was able to delve into the nontrivial ways epidemics spread through social networks. This modeling approach simplified the realworld dynamics into more analytically tractable forms, allowing the researchers to capture the essence of contact network structures and their crucial role in transmitting infectious diseases. The primary objective of this study was to identify new pathways for academic exploration and offer valuable perspectives that can enhance public health policies and epidemic response strategies. Ultimately, this work seeks to contribute to a better understanding of epidemic dynamics by bridging knowledge gaps and fostering a more resilient response to public health challenges in the face of complex human interactions.
  • Item
    Machine Learning for Structure Search of Ligand-protected Nanoclusters
    (Aalto University, 2024) Fang, Lincan; Xi, Chen, Prof., Lanzhou University, China; Teknillisen fysiikan laitos; Department of Applied Physics; Perustieteiden korkeakoulu; School of Science; Rinke, Patrick, Prof., Aalto University, Department of Applied Physics, Finland
    Understanding the atomic structures of ligand-protected nanoclusters is essential for their application in various fields. These structures not only determine the physical and chemical properties of ligand-protected nanoclusters but also play a crucial role in their stability and reactivity. Knowing the precise atomic structures allows us to tailor nanoclusters for specific functions. However, because of the extraordinarily high dimensionality of the search space which encompasses an exceptionally large number of all potential structures, it is difficult to use quantum mechanical methods, such as the density functional theory, to find the low-energy structures of ligand-protected nanoclusters. On this point, the structure search of ligand-protected nanoclusters could be more efficient and accurate by utilizing machine learning methods. In this dissertation, I developed machine learning methods to search the atomic structures of ligand-protected nanoclusters by decomposing the problem into three steps. For the first step, I developed a molecular conformer search procedure based on Bayesian optimization to search the structures of isolated molecules. Using four amino acids as examples, I showed that the procedure is both efficient and accurate. For the second step, I modified the procedure to search the structures of a single ligand on a nanocluster. I also developed and tested strategies to avoid steric clashes between a ligand and cluster parts. Moreover, I tested and demonstrated our modified procedure by searching structures for a cysteine molecule on a well-studied gold-thiolate cluster. As a result, I found that cysteine conformers in a cluster inherit the hydrogen bond types from isolated conformers, while the energy rankings and spacings between the conformers are reordered. In the final step, I applied a machine learning method based on kernel rigid regression (KRR) models to relax the structures of ligand-protected nanoclusters. Moreover, I used an active learning workflow to enhance the relaxation performance of the KRR models. To test and demonstrate our method, I applied it to search structures of Au25(Cys)18 -. We found that the low-energy structures with IItype hydrogen bonds (OH- -N, OH from trans-COOH and N from NH2) are dominant and the different configurations of the ligand layer indeed influence the properties of the clusters.
  • Item
    Interview Methodologies for Addressing Autobiographical Ruptures and (Re)Constructing Life Narratives - A Longitudinal Study Among Prostate Cancer Patients and Their Spouses
    (Aalto University, 2024) Talsi, Riikka; Saarinen, Esa, Prof. Emeritus, Aalto University, Department of Industrial Engineering and Management Science, Finland; Laitila, Aarno, Senior Lecturer, University of Jyväskylä, Finland; Joensuu, Timo, Prof., Docrates Cancer Center, Finland; Tuotantotalouden laitos; Department of Industrial Engineering and Management; Perustieteiden korkeakoulu; School of Science; Vuori, Natalia, Asst. Prof., Aalto University, Department of Industrial Engineering and Management, Finland
    Major life changes may cause an autobiographical rupture and the need to work on one's narrative identity. In this dissertation, I developed new interview methodologies to work through an auto-biographical rupture experience and to support the (re)construction of life narratives. Ten newly diagnosed prostate cancer patients, five with their spouses, participated in the study. Each patient/ couple participated in five narrative in-depth interviews during cancer treatment and later in a follow-up interview. The interview series offered the participants several ways to talk about their cancer experience and life, and to re-evaluate and reconstruct what they had previously said. In the Momentary Key Metaphor methodology, the interviewee is asked to describe the illness experience using metaphors and to reflect on their personal meaning. In the follow-up interview, the metaphors are returned to the interviewee for retrospective review. The metaphors provided the participants with a tool to explore, summarize, and reflect on a complex, often contradictory experience at different stages of its development. One key finding was the polyphonic nature of the metaphors. The construction of polyphonic metaphors helped the participants work with emotions, tolerate uncertainty, and deal with different aspects of their experience. The couple context introduced the voice of the spouse and highlighted the mutuality in the couple's dyadic coping process. At the beginning, when uncertainty was high, the spouses used metaphors to create hope. As the patients' agency became stronger, the spouses were able to express their own vulnerability. In the Clip Approach methodology, the interviewee's narration is reflected back through visual artifacts, "the Clips," that allow the interviewee to re-enter their cancer experience and life, and re-construct their narratives concerning them. The Clips returned the narration in a tangible, reconstruable form in the narrator's own voice. They supported participants in developing self-observation and helped them move from the object position of a serious illness to the subject position of an active agent. The construction of a life story through the use of the Clips supported autobiographical reasoning and helped to explore and re-evaluate meanings, build a bridge between the past and the future and embed the cancer experience as part of the participant's life narrative. In the dissertation, I propose a series of interviews as a patient-centered guidance and counseling intervention for cancer care. The methodologies can be used to address and explore autobiographical ruptures in a variety of life-changing situations, and they offer encouraging potential for low-threshold psychosocial support interventions in a range of application areas. They may also provide new tools for existing contexts. The Clip Approach suits environments with time for collaborative work, like psychotherapies. The Momentary Key Metaphor can be used as a dialogic opening even in shorter encounters, for example during medical appointments in a longer treatment relationship.
  • Item
    Applicability and Robustness of Deep Learning in Healthcare
    (Aalto University, 2024) Sahlsten, Jaakko; Kaski, Kimmo, Prof., Aalto University, Department of Computer Science, Finland; Tietotekniikan laitos; Department of Computer Science; Perustieteiden korkeakoulu; School of Science; Lampinen, Jouko, Prof., Aalto University, Department of Computer Science, Finland
    The worldwide population is aging causing an increased demand for healthcare, motivating a goal to reduce the burden of health professionals to maintain the expected level of care. Deep learning (DL)-based methods, especially convolutional neural networks (CNNs), have achieved state-of-the-art performance in various classification and segmentation tasks on imaging data. Thus, there is an interest in applying these methods to automate routine, laborious or time-consuming clinical tasks based on medical imaging. However, the conventional DL approaches may not be trustworthy to be used in healthcare due to limited explainability combined with overconfidence and their sensitivity to distribution shifts. In this thesis, the applicability of DL approaches in healthcare are investigated in the clinical tasks of screening and medical image segmentation. The approaches are evaluated for robustness to distribution shifts with in-distribution and out-of-distribution datasets including other imaging centers, devices, and under defacing techniques. In order to improve the lack of explainability and overconfidence, approximate Bayesian neural networks with novel uncertainty measures are applied to the tasks and systematically evaluated in terms of performance and uncertainty quantification. The deep learning paradigm and its practical usage in the investigated medical imaging tasks is first introduced. The following part describes uncertainty quantification in deep learning, its downstream utilization in clinical workflow, and the current approaches of approximate Bayesian deep learning. The next part includes the summary for the included publications and related works. The last part includes the conclusion and discussion about the analysis, its limitations, and proposed future research to improve the trustworthiness and applicability of deep learning techniques for imaging in healthcare. The publications demonstrated that CNN-based DL methods have clinically acceptable performance in the evaluated tasks using in-distribution data. However, the robustness to distribution shift varied depending on the task such as robustness to other imaging devices but sensitivity to defacing in segmentation. In terms of explainability and overconfidence, the approximate Bayesian deep learning and the novel uncertainty measures demonstrated improved utility of uncertainty in comparison to conventional approaches in both tasks.
  • Item
    Probabilistic Methods for Predictive Maintenance and Beyond: Graph and Human-in-the-Loop Machine Learning
    (Aalto University, 2024) Nikitin, Alexander; Tietotekniikan laitos; Department of Computer Science; Probabilistic Machine Learning; Perustieteiden korkeakoulu; School of Science; Kaski, Samuel, Prof., Aalto University, Department of Computer Science, Finland and The University of Manchester, United Kingdom
    Probabilistic methods are key tools for machine learning problems. Even so, there remain many applications where they cannot be applied due to their limitations. These limitations may include the lack of methods for a particular data format (e.g., manifolds, texts, or graphs), data unavailability, or the inability to work collaboratively with human experts. Inspired by problems in predictive maintenance (PdM), this thesis introduces a set of machine learning solutions that are more generally applicable. It begins with applied tasks in cable networks, data centers, and other telecom applications and indicates the crucial limitations of current approaches: the absence of (i) probabilistic methods for spatio-temporal graph problems, (ii) practical human-in-the-loop methods that learn from detailed domain experts' feedback, and (iii) systems for synthetic temporal data creation that enable secure sharing of sensitive data between parties. Moreover, even if such methods become available, it is important to describe how those methods can be used in an end-to-end system for predictive maintenance covering both the modeling and operations sides. This thesis analyses and resolves these issues. The first issue, the lack of probabilistic methods for graph and spatio-temporal graph data, was resolved by connecting graph kernels with stochastic partial differential equations (SPDEs). This method results in a variety of kernels suitable for machine learning problems on graphs, including Mat\'ern, stochastic heat, and stochastic wave graph kernels. The second issue, the lack of human-in-the-loop methods with domain experts' explicit feedback, was resolved by developing a decision rule elicitation mechanism and its domain adaptation properties. The method is grounded in human decision-making mechanisms and has been tested in several user studies. It leads to a simple yet effective method for working with domain experts. Next, synthetic data generation was resolved by introducing an open-source software framework called TSGM. This framework effectively generates synthetic time series data and provides a toolkit for evaluation. This work also examined the various approaches to the generation and evaluation of synthetic data. Finally, the methods proposed in the thesis resulted in successful real-world implementations tested on several large-scale cases with our industrial partner Elisa Oyj. Furthermore, those implementations led to five submitted patents, one of which has already been granted. This thesis discusses the aforementioned results, places them into a broader perspective, and provides possible avenues for future research.
  • Item
    Investigating the white matter structure of the sensorimotor system in children with cerebral palsy
    (Aalto University, 2024) Jaatela, Julia; Piitulainen, Harri, Prof., University of Jyväskylä, Finland; Neurotieteen ja lääketieteellisen tekniikan laitos; Department of Neuroscience and Biomedical Engineering; Perustieteiden korkeakoulu; School of Science; Parkkonen, Lauri, Prof., Aalto University, Department of Neuroscience and Biomedical Engineering, Finland
    Cerebral palsy (CP) is the leading motor disorder in childhood, primarily caused by a brain insult occurring before birth. CP is typically characterized by spasticity, affecting one side of the body (hemiplegia) or both sides, predominantly the lower limbs (diplegia). The advancements in diffusion-weighted magnetic resonance imaging (dMRI) and tractography, have allowed the identification of altered white matter structure of the brain in CP. However, our understanding of how these alterations differ between hemiplegic and diplegic subtypes, and their relationship with sensorimotor deficits, such as balance, gait and manual dexterity, remains limited. In this thesis, we investigated a cohort of children and adolescents aged 10–18 years diagnosed with hemiplegic (N = 16) or diplegic CP (N = 11) alongside their typically developed peers (N = 31). Using tractography, we investigated the interhemispheric commissural pathways, i.e. corpus callosum, and thalamocortical pathways connecting to the representational areas of the upper and lower limbs. To address the cortical abnormalities seen in CP and enhance the functional relevance of the studied tracts, we introduced a novel seeding approach using proprioceptive simulation (passive movement) of the limbs together with functional neuroimaging. The derived dMRI metrics were compared between the three groups, and their association with behavioral measures was investigated. Our results showed significant alterations in the diffusion properties of the investigated pathways between children with and without CP, indicating changes in the axonal organization. Specifically, participants with hemiplegic CP seemed to have more severe structural changes that were relatively localized when compared to those with diplegic CP. The white matter involvement reflected, to some extent, the topographic presentation of the functional deficit. While we observed some associations between the dMRI-metrics and sensorimotor function, they were weak, and the directionality was non-conclusive, underscoring the complexity of the structure-function relationships. The dMRI holds promising potential as an objective tool for guiding the diagnosis and treatment of CP in the future. By highlighting the differences in location and severity of white matter alterations between hemiplegic and diplegic CP, our findings contribute significantly to the existing literature. Further, our research emphasizes the importance of along-tract analysis and specific outlining of investigated tracts in future studies on both CP and typical development. Research on CP not only enhances our understanding of the disorder itself but also sheds light on the development and plasticity of the human sensorimotor system.
  • Item
    Droplet friction on heterogeneous surfaces
    (Aalto University, 2024) Lepikko, Sakari; Teknillisen fysiikan laitos; Department of Applied Physics; Soft Matter and Wetting; Perustieteiden korkeakoulu; School of Science; Ras, Robin, Prof., Aalto University, Department of Applied Physics, Finland
    The ability to control friction between a solid and a liquid is becoming more and more important in various existing applications as well as in novel ones. However, the mechanisms behind this liquid-solid friction are not yet sufficiently understood. This thesis compiles research performed in three publications where origin of the liquid-solid friction of water droplets is examined for various surfaces with properties ranging from hydrophilic to superhydrophobic and from a regular surface structure to a stochastic one. Publication I examines how molecular level heterogeneity of a surface affects the contact line friction between the surface and water. This is performed by preparing and characterizing self assembled monolayers with varying level of molecular coverages of hydrophobic alkyl tails on hydrophilic silicon wafer substrates. The results show that the low- and high-coverage surfaces with least chemical heterogeneity have the lowest friction while the intermediate-coverage surfaces with most heterogeneity have the highest friction. Publication II focuses on the relation of the contact line friction and the liquid-solid contact fraction of superhydrophobic surfaces. The friction is shown to scale with the contact fraction, being lowest with the lowest contact fraction, and a mathematical model is provided to describe this relation. The model works over wide ranges of friction and contact fraction values, both extending almost over three orders of magnitude. Another important message is that conical microstructures can be used to create surfaces with an extremely low liquid-solid contact fraction that results in an extremely low contact line friction. Publication III explores how superhydrophobic surfaces with stochastic roughness in the nanoand micrometre scales affect the liquid-solid friction. The wetting characterization shows that the friction is time dependent such that static droplets have time to adapt to the surface roughness while moving droplets do not have such time. This adaptation increases the liquid-solid contact fraction, which causes the increased contact line friction. Effectively, this creates a static friction barrier that pins static droplets to the surface but does not restrict the movement of already mobiledroplets. The obtained results of Publications I-III help determining critical surface parameters when designing functional surfaces for applications where low friction between a solid and a liquid is needed.
  • Item
    Transport of interacting particles through a flat Bloch band: superconductivity and all-optical switching
    (Aalto University, 2024) Pyykkönen, Ville; Peotta, Sebastiano, Dr., Aalto University, Department of Applied Physics, Finland; Salerno, Grazia, Dr., Aalto University, Department of Applied Physics, Finland; Teknillisen fysiikan laitos; Department of Applied Physics; Quantum Dynamics; Perustieteiden korkeakoulu; School of Science; Törmä, Päivi, Prof., Aalto University, Department of Applied Physics, Finland
    In lattice models, destructive interference can lead to formation of localized states, which results in Bloch bands with a constant dispersion, called flat bands. Flat-band systems give a promising platform for high-temperature superconductivity. They also allow potential applications for signal control in electronics and photonics. In this dissertation, we study transport features of interacting fermions and bosons through systems possessing compactly localized flat-band states. We propose and investigate a two-terminal transport setup for studying these states in the context of flat-band superconductivity. Also, we propose a switching concept for photons based on localized states and their sensitivity to interactions. The dissertation consists of an introductory part and three publications, referred to as I, II, and III. The introductory part discusses the essential theoretical background, giving a brief review of the main results, their implications and outlook. The topics covered include a general introduction to flat Bloch bands, superconductive transport, and non-equilibrium Green's functions, with applications to two-terminal transport.     In Publication I, we consider a two-terminal setup for studying localized flat-band states with interaction. Focusing on equilibrium transport via the Josephson effect, we show that connecting superconducting leads to the system allows a supercurrent to flow through flat-band states with on-site Fermi-Hubbard interaction. The critical current and critical temperature are found to be linear in interaction strength, a salient feature of the flat bands. We also consider a potential realization of the system in ultracold gases. In Publication II, we consider non-equilibrium superconductive transport through flat-band states with on-site Fermi-Hubbard interaction in the setup proposed in Publication I. The interactions are considered with a mean-field approximation. We solve the stationary state transport by the method of non-equilibrium Green's functions, showing that normal single-particle transport and transport via Andreev reflection and multiple Andreev reflection, involving quasiparticles, are quenched at the flat band. On the other hand, the AC Josephson effect of Cooper pairs is allowed. Hence, we find that pair transport through flat-band states is allowed while single quasiparticles remain localized. In Publication III, we propose an all-optical switching concept based on localized states. We demonstrate the concept with simple systems that have on-site Hubbard interaction. We show that the system allows switching a single photon by a single-photon control pulse, which is the fundamental quantum limit of minimal switching energy. Furthermore, the switching is allowed for arbitrarily small interaction. We also discuss experimental platforms for realizing the switch.
  • Item
    Particle scattering in magnetised plasmas: a theoretical and numerical approach
    (Aalto University, 2024) Iorio, Riccardo Nicolò; Hirvijoki, Eero, Dr., Aalto University, Department of Mechanical Engineering, Finland; Kiviniemi, Timo, Dr., Aalto University, Department of Applied Physics, Finland; Teknillisen fysiikan laitos; Department of Applied Physics; Perustieteiden korkeakoulu; School of Science; Groth, Mathias, Prof., Aalto University, Department of Applied Physics, Finland
    The intricate dynamics of charged particles within plasmas are mainly shaped by their collisional interactions. As a result, it is crucial to address these phenomena through both theoretical and numerical approaches. In pursuit of this objective, this work embarks on reviewing the formal derivation of the Vlasov equation, followed by an extensive exploration of Coulomb scattering, elucidating the Landau collision integral and its underlying characteristics. Furthermore, we delve into the nonconventional neoclassical theory for toroidal systems, providing the theoretical framework for the subsequent numerical findings. Utilizing the ELMFIRE code, gyrokinetic simulations employ a discrete Landau collision integral, ensuring the conservation of energy and momentum. Tailored to conservation laws, a specific binary collision model provides valuable insights into variations in impurity density arising from steep gradients in density and temperature profiles. The analysis compares Landreman-Fülöp- Guszejnov model's theory with neoclassical predictions and ELMFIRE data. Remarkably, within the analytical theory's validity, numerical agreement is 5-10%, especially for δ<0.4 with low charge numbers. Yet, within the pedestal region, the Landreman-Fülöp-Guszejnov framework may not be directly applicable due to pronounced gradients. Furthermore, a novel analysis explores the correlation between turbulent transport and the radial electric field. Using Lower Hybrid (LH) heating operator in an FT-2 tokamak at off-axis and onaxis reveals heightened turbulence at r/a=0.55 during a 70μs simulation. Turbulence induces noticeable fluctuations in the radial electric field profile, with strong high-shearing flow in the former and neoclassical dominance in the latter. These findings align with prior research, suggesting a robust shearing phenomenon, reinstating transport equilibrium. In conclusion, to enhance the central theme of this dissertation, we investigate the formal derivation of a collisional bracket from the Landau collision integral using the metriplectic bracket formulation for dissipative systems. This theoretical framework is then applied to the guiding center Vlasov-Maxwell-Landau model, resulting in a specific collisional bracket that ensures energy and momentum conservation. The implications of this finding are explored within broader frameworks, including the electromagnetic gyrokinetic case, offering a theoretical culmination to this dissertation.
  • Item
    Efficient and trustworthy methods for knowledge discovery
    (Aalto University, 2023) Ciaperoni, Martino; Aslay, Cigdem, Prof., Aarhus University, Denmark; Tietotekniikan laitos; Department of Computer Science; Data Mining; Perustieteiden korkeakoulu; School of Science; Gionis, Aristides, Prof., KTH Royal Institute of Technology, Sweden and Aalto University, Department of Computer Science, Finland
    Data are building blocks to information and, subsequently, they are vital input to knowledge. Today, in the midst of the digital era, vast quantities of highly-complex data are being collected and processed at an unprecedented scale. This abundance of data has highlighted the importance of efficient and effective knowledge-discovery algorithms to identify patterns hidden in the data with the ultimate aim of uncovering valuable knowledge and shape our understanding of the world around us. To capitalize on the opportunities offered by massive amounts of data as well as modern computing power, for many years, research in knowledge discovery and related areas has introduced algorithms that are increasingly efficient and effective, but also more and more opaque and unpredictable. Recently, growing interest in the ethical dimensions of algorithms has drawn attention to the limitations of opaque algorithms and has emphasized a need for trustworthy algorithms particularly when such algorithms are used to support high-stakes decision making. In order to be trustworthy, algorithms should solve a clearly defined problem via a clear sequence of instructions, they should not be utterly unsuccessful in any particular case and they should be easy to understand and interpret for humans so that no harmful biases can be hidden. In this thesis, we pursue the goal of developing novel knowledge-discovery algorithmic methods that are not only highly efficient to face the challenges and opportunities posed by modern data, but also trustworthy. In particular, we propose efficient and trustworthy methods for a collection of popular knowledgediscovery tasks. First, we consider tasks of exact inference in Bayesian networks and hidden Markov models. Trustworthy approaches for such tasks exist. However, their applicability may be severely limited by time or memory requirements. Therefore, we propose novel methods to reduce the time or memory resources that are needed by existing approaches for the considered exact inference tasks. Beside exact inference tasks, we also consider two different knowledge-discovery tasks that arise naturally in modern data: multi-label classification and community search in temporal graphs. Regarding multi-label classification, we propose an efficient and accurate rule-based multi-label classifier that drastically improves upon the interpretability of existing solutions. For community search in temporal graphs, we formalise the task for the first time, and we propose a solution that guarantees high efficiency and interpretability. In designing knowledge-discovery methods, we often rely on existing database-management and probabilistic methods. Methods for database management are valuable to address the large dimension and high complexity of modern data, while probabilistic methods are essential to methodologically handle uncertainty in the data.
  • Item
    Bayesian Optimal Experimental Design in Imaging
    (Aalto University, 2023) Puska, Juha-Pekka; Hyvönen, Nuutti, Prof., Aalto University, Department of Mathematics and Systems Analysis, Finand; Matematiikan ja systeemianalyysin laitos; Department of Mathematics and Systems Analysis; Perustieteiden korkeakoulu; School of Science; Hyvönen, Nuutti, Prof., Aalto University, Department of Mathematics and Systems Analysis, Finand
    An inverse problem is defined as a problem that violates one of the classical criteria of a well posed problem: a solution exists, is unique, and is continuous with respect to the data in some reasonable topology. A problem that is not well posed is called illposed, and the development of tools to tackle illposed problems is the goal of the field of inverse problems research. In imaging, illposedness is often an inevitable consequence of the high dimension of the unknown, compared with the measurement data. In an imaging problem, one aims to reconstruct the spatial two- or three-dimensional structure of an object of interest, leading to unknown parameters in the hundreds of thousands or beyond, while the dimension of the measurement data is determined by the number of sensors, and thus limited by physical constraints to values often at least an order of magnitude lower. Another consequence of the high dimensionality of the problem is the computational cost involved in the computations. In imaging problems, there is also usually a cost involved in acquiring data, and thus one would naturally want to minimize the amount of data collection required. One tool for this is optimal experimental design, where one aims to perform the experiment in a way as to maximize the value of the data obtained. The challenge of this however, is that the search for this optimal design usually leads to a computationally challenging problem, whose size is dependent on the dimension of both the data and the unknown. Overcoming this difficulty is the main objective of this thesis. The problem can be tackled by using Gaussian approximations in the formulation of the imaging problem, which leads to practical solution formulas for the quantities of interest. In this thesis, tools are developed to enable the efficient computation of expected utilities for certain measurement designs, particularily in sequential imaging problems and for non-Gaussian prior models. Additionally, these tools are applied to medical imaging and astronomy. 
  • Item
    Designing artificial moiré van der Waals topological superconductivity
    (Aalto University, 2023) Khosravian, Maryam; Liljeroth, Peter, Prof., Aalto University, Department of Applied Physics, Finland; Teknillisen fysiikan laitos; Department of Applied Physics; Correlated Quantum Materials (CQM) group; Perustieteiden korkeakoulu; School of Science; Lado, Jose, Prof., Aalto University, Department of Applied Physics, Finland
    The study of topological superconductivity is a promising field in condensed matter physics that has exciting possibilities for the emergence of exotic quantum phenomena and topological quantum computing. In this thesis, we investigate various platforms for engineering topological superconductivity, focusing on the role of quasiperiodicity and van der Waals materials featuring moiré patterns. The first scheme concentrates on quasiperiodic systems and their capability to generate robust spin-triplet superconducting pairings through coexisting orders, establishing a new strategy for engineering unconventional superconductivity. In the second scheme, we focus on specific van der Waals heterostructures showing moiré patterns. We examine the role of impurities in designer topological moiré superconductors that combine van der Waals magnetic and superconducting materials, focusing on the interplay between atomic and moiré length scales within these artificial moiré systems explored with conventional tools and machine learning. In the third scheme, we focus on hybrid van der Waals heterostructures based on twisted graphene bilayers, magnets, and superconductors, establishing their potential as a versatile platform for engineering artificial topological  superconductivity. Our results showcase the potential of designer platforms, such as quasiperiodicity and moiré-patterned van der Waals materials, to harness and manipulate these intriguing quantum states. Our findings establish new strategies for developing quantum technologies based on topological superconducting quantum materials and further enrich our understanding of exotic quantum matter.
  • Item
    Parameterized Approximation Results for Clustering and Graph Packing Problems
    (Aalto University, 2023) Gadekar, Ameet; Brzuska, Chris, Prof., Aalto University, Department of Computer Science, Finland; Tietotekniikan laitos; Department of Computer Science; Theoretical Computer Science; Perustieteiden korkeakoulu; School of Science; Chalermsook, Parinya, Prof., Aalto University, Department of Computer Science, Finland
    The domains of clustering and graph packing have been the focus of extensive research across multiple disciplines, including optimization, machine learning, data mining, computational geometry, and operations research. Many central problems in these domains are known to be NP-hard, prompting the exploration of approximation algorithms and parameterized algorithms as possible approaches, among others. In recent times, the paradigm of parameterized approximation algorithms, which strike a balance between approximation and polynomial-time computability on instances with small parameters, has gained renewed attention.  This thesis comprises two distinct parts. In Part I, we design parameterized approximation algorithms for several clustering problems, significantly advancing the state of the art. In the center based k-clustering problem, which includes the classical problems of k-median, k-means, and k-center, we are given a point set P and the goal is to find a k-partition (clusters) of P along with a representative (center) for each partition to minimize certain clustering objective that is a function of the distance vector - the vector of distances between the centers and the points in the corresponding clusters. In the Norm k-Clustering problem, the clustering objective is a monotone norm of the distance vector. This objective is quite general and captures essentially all the clustering objectives studied so far. In this thesis, we design a novel and simple Efficient Parameterized Approximation Schemes (EPAS) framework for Norm k-Clustering in several metric spaces. This result unifies several existing EPASes that are known to be conceptually different. Moreover, our framework resolves many of the open problems related to advanced objectives, including modern constraints on fairness, robustness, and diversity. A notable contribution of this work is a new combinatorial measure of a metric space, which we call Scatter Dimension, that enables designing EPAS that is oblivious to the underlying metric space. Additionally, we address other clustering problems, namely Robust (k-z)-Clustering and Diversity-aware k-Median, and design tight parameterized approximation algorithms for them.  Part II adopts a complementary approach, focusing on lower bounds for graph packing problems. In particular, we consider a notoriously hard problem called Set Packing and establish a parameterized dichotomy for the problem. A novel conceptual contribution to this dichotomy is the notion of compact instances, which remain challenging to solve despite their small size. Furthermore, we explore the connection between the approximating maximum independent set problem in k-claw-free graphs and several convex relaxations.
  • Item
    Harnessing Quantum Resources in Superconducting Devices for Computing and Sensing
    (Aalto University, 2024) Perelshtein, Mikhail; Teknillisen fysiikan laitos; Department of Applied Physics; NANO group (Quantum Circuits and Correlations); Perustieteiden korkeakoulu; School of Science; Hakonen, Pertti, Prof., Aalto University School of Science, Department of Applied Physics, Finland
    Quantum resources play a pivotal role in the emerging field of quantum technologies, underpinning the development of quantum computers, communication systems, and sensing devices with transformative capabilities beyond classical counterparts. In this thesis, key quantum resources, such as entanglement and coherency, are investigated with an emphasis on superconducting metamaterials, including Josephson Parametric Amplifiers, Travelling Wave Parametric Amplifiers, and artificial transmon atoms. The presented thesis strives to explore, unify, and harness the quantum resources in quantum sensing and computing tasks. The initial cluster of studies places a spotlight on the experimental preparation of quantum states enriched with quantum resources. The thesis presents the generation of multipartite quantum entanglement using innovative pump tone techniques applied to the Josephson parametric system, expanding the possibilities for microwave control of the entanglement structure and demonstrating for the first time genuine entanglement between four microwave modes. The generation of frequency-entangled photons with record-breaking 4 GHz frequency separation is presented for distributed Josephson metamaterial, highlighting the practical implications of entangled microwave photons in broadband quantum information processing. Besides, the thesis presents the protocol for preparation of multi-qubit states with target amplitudes, which features polylogarithmic scaling in the number of encoded parameters. The protocol demonstrates its effectiveness with large-scale circuits up to 100 qubits, which are studied numerically. The presented work advances phase estimation protocols for quantum sensing, optimizing resource utilization and sensitivity, particularly in the realm of magnetic field measurements. It further explores the effectiveness of separable and entangled states for magnetometry by conducting experiments on existing quantum hardware through cloud-based IBM quantum systems. A novel sensing algorithm for multi-level artificial atoms emerges from this exploration, designed to maximize the utilization of available quantum resources, which is numerically investigated. Finally, the thesis presents a study of the hybrid quantum algorithm for solving large linear systems of equations, showcases the current state of intermediate-scale quantum computers, and proposes the benchmark for future hardware developments. It introduces a classification scheme based on entanglement structure and successfully implements a record-breaking 217-dimensional problem on IBM quantum processors.
  • Item
    Critical study of contact angle goniometry
    (Aalto University, 2024) Huhtamäki, Tommi; Teknillisen fysiikan laitos; Department of Applied Physics; Soft Matter and Wetting; Perustieteiden korkeakoulu; School of Science; Ras, Robin, Prof., Aalto University, Department of Applied Physics, Finland
    In this thesis contact angle goniometry as a characterization method for solid surface wettability is critically examined. Contact angle goniometry is a powerful and versatile tool for measuring wetting properties, facilitating the study of microscale surface properties by macroscopic an optical method. The apparent ease which with contact angles can be measured have ensured its position as the golden standard of wetting characterization. Measuring meaningful contact angles and interpreting the data correctly is far from simple, however. Real solid surfaces exhibit a range of stable contact angles. Collecting meaningful data requires knowledge to recognize the contact angles which can be reproducibly measured, on how to perform the measurements in a reliable manner and on interpretation of the results. Publication I provides a method for reliable measurement of the receding contact angle. The validity of the method is evaluated both theoretically, and by a wide range of experimental evidence. Publication II proposes a growth model for the synthesis of silicone nanofilaments - a class of superhydrophobic 1-dimensional polysiloxane nanostructures. Theoretical model for pressureinduced film expansion is provided. Publication III introduces a protocol for contact angle measurements that can be applied to a wide variety of samples. Instructions on minimizing both systematic and random errors are provided, along with troubleshooting for most common problems encountered. Publication IV quantifies the error in contact angle measurement caused by misplacement of the baseline - the line between the solid and liquid/gas in the 2D-images analyzed. Special emphasis is given on the error for superhydrophobic surfaces. Publication V expands on publications IV by quantifying the error caused by the optical system used in contact angle goniometry. User error of contact angle measurements is also meas-ured.
  • Item
    Sample-efficient inference for agent-based cognitive models and other computationally intensive simulators
    (Aalto University, 2023) Aushev, Alexander; Tietotekniikan laitos; Department of Computer Science; Probabilistic Machine Learning; Perustieteiden korkeakoulu; School of Science; Kaski, Samuel, Aalto University, Finland and The University of Manchester, United Kingdom
    In recent years, simulator models have become increasingly popular in many scientific domains, such as epidemiology, cosmology, and behavioural sciences. Since simulators often do not have tractable likelihoods, which are either too costly to evaluate or not available, the field needs to resort to likelihood-free inference (LFI), which uses forward simulations instead. With the development of more complex simulators, traditional LFI methods become unfeasible as the cost of simulations significantly increases. This thesis deals with three challenges that arise in the context of computationally heavy simulators and for which the existing LFI methods, such as approximate Bayesian computation, synthetic likelihood, or neural density estimation approaches, are inadequate since they require a large number of simulations. The first challenge is modelling complex simulator noise, which influences the accuracy of LFI methods and becomes problematic when simulations are computationally costly. The existing methods either oversimplify the noise (e.g., by assuming it to be Gaussian) or require an infeasible number of simulations to accurately model it. We show how to handle multimodal, non-stationary, and heteroscedastic noise distributions in LFI while also assuming a small simulation budget. For this, we adopt deep Gaussian process surrogates in Bayesian Optimisation (BO), along with novel quantile-based multimodal-capable modifications for the acquisition function and posterior extraction procedures. Another challenge for modern LFI approaches occurs when they are applied to time-series settings, as these methods either need an accurate model of transition dynamics available or always assume it to be linear. We propose a way of estimating the unknown transition dynamics for state predictions in simulator-based dynamical systems, which greatly reduces the required simulation budget and also enables time-series prediction. Our proposed approach uses a multi-objective surrogate for LFI and a semi-parametric model for the transition dynamics. Finally, we significantly reduce the time required to select agent-based cognitive models with limited experimental designs. The previous methods have primarily focused on either model selection or parameter estimation, while we achieve both in a fraction of the time. This is accomplished through a novel simulator-based utility objective for selecting designs in BO and a LFI approximation of model marginal likelihood for model selection. This new method is needed for developing and verifying computational cognitive theories, which often lack tractable likelihoods.
  • Item
    Interpretable artificial neural networks for fMRI data classification
    (Aalto University, 2023) Gotsopoulos, Athanasios; Sams, Mikko, Prof., Aalto University, Department of Neuroscience and Biomedical Engineering, Finland; Neurotieteen ja lääketieteellisen tekniikan laitos; Department of Neuroscience and Biomedical Engineering; Brain & Mind Laboratory; Perustieteiden korkeakoulu; School of Science; Lampinen, Jouko, Prof., Aalto University, Department of Computer Science, Finland
    Functional magnetic resonance imaging (fMRI) technology allows non-invasive measurement of neuronal activity in the human brain with a combination of reasonable temporal and fine spatial resolution. Recently, multivariate methods have attracted attention in fMRI data analysis to study task-related activation patterns. Concurrent research in the field of machine learning has led to the establishment of inherently multivariate computational graphs that facilitate efficient, robust and interpretable classification of fMRI data. Here we studied methods for classification of fMRI data based on neural networks. In particular, we focused on techniques that assess the contribution of different brain regions to the classification result, referred to as "importance maps" and proposed novel neuroscientifically motivated architectures. In the first study, we successfully classified basic emotions from fMRI data, elicited by short movies and mental imagery, generating whole brain importance maps indicating the contribution of individual voxels to the classification result. The second study provided a comparison of importance extraction methods and their reproducibility, applied to both simulated and real data sets, revealing patterns that do not convey significant univariate information. The third study examined the effect of distractors in visual imagery using classification methods and importance map extraction, identifying robust activation patterns related to shape imagery and a visual distractor in object-selective lateral extrastriate cortex at the junction of left occipital, temporal and parietal lobes. The fourth study examined the use of anatomically driven topologies based on spatial information. In particular, the addition of layers motivated by voxel proximity and brain atlases to the model, led to an increase in the classification accuracy and produced smoother and more interpretable importance maps. The purpose of this thesis is to showcase machine learning techniques specifically designed for analyzing neuroscience data. This work aims to motivate further research towards the use of machine learning as a means to gain a better understanding of the human brain.