http://hdl.handle.net/10900/42133 2026-06-12T07:14:49Z 2026-06-12T07:14:49Z Kremer, Heiner Stephan http://hdl.handle.net/10900/180527 2026-06-10T01:04:30Z 2026-06-09T00:00:00Z

Empirical Likelihood Estimators for Robust and Causal Learning Kremer, Heiner Stephan Some of the central problems in robust and causal machine learning, including learning under covariate shifts and instrumental variable regression, can be expressed as conditional moment restrictions (CMR). By restricting the conditional expectation of a signed error metric, models identified via CMR exhibit robustness against shifts in the distribution of the conditioning variable. In practice, this generally results in an ill-posed problem, as it requires the solution of an over-identified infinite-dimensional system of equations. For the unconditional case, empirical likelihood estimators have emerged as general and powerful tools to address over-identified moment restriction problems. These methods learn a model along with an approximation of the population distribution by means of minimizing a φ-divergence constrained by the moment restrictions. The main goal of this work is to advance the state-of-the art in CMR estimation by extending and refining the idea of empirical likelihood estimation in several directions. First, we generalize the classical framework to conditional moment restrictions using a functional formulation, that leverages modern machine learning models. Then, we extend the principle to alternative distributional distance notions based on kernel methods and optimal transport. The resulting estimators exhibit superior small sample properties and robustness against data corruptions at training time and adversarial attacks at test time, respectively. Finally, drawing inspiration from the close relation between empirical likelihood estimation and distributionally robust optimization (DRO), we provide an application of kernel-based DRO on chance-constrained programming.

2026-06-09T00:00:00Z Vanegas Torres, Carlos Alberto http://hdl.handle.net/10900/180503 2026-06-10T01:04:29Z 2026-06-09T00:00:00Z

Vpr-Controlled Manipulation of T Cell Physiology: NF-AT Activation and Protein Degradation as Key Mechanisms for HIV-1 Pathogenesis Vanegas Torres, Carlos Alberto Human immunodeficiency virus type 1 (HIV-1), the lentiviral pathogen behind the global AIDS pandemic, preferentially infects CD4+ T lymphocytes, leading to their progressive depletion via both direct viral cytotoxicity and through increased rates of apoptosis. To achieve full pathogenicity in vivo, HIV-1 encodes multiple accessory proteins, most of which play defined roles at various steps of the viral replication cycle. In contrast, the 96-amino acid Viral Protein R (Vpr) is implicated in disrupting host cell physiology through a variety of mechanisms, such as facilitating the nuclear import of viral pre-integration complexes, as well as significantly boosting viral production by enhancing the transcriptional activity of viral LTRs. Further, Vpr is actively encapsidated into HIV-1 virions, allowing its direct delivery into host cells upon de novo infection. Collectively, these characteristics epitomize Vpr as a crucial supporting element in the establishment of a productive HIV-1 infection. Nevertheless, multiple gaps exist in the understanding of the mechanisms whereby Vpr allows HIV-1 to exert control over its host cell at various organizational levels, and many studies still fail to answer these questions in physiologically relevant models, such as donor-derived CD4+ T lymphocytes. To address these issues, HIV-1 infection assays employing inhibitors for various signaling pathways were performed on T cell-derived models and primary CD4+ T cells alike, focusing on Vpr’s role in the induction of NF-AT signaling. Consecutively, a thorough bioinformatic analysis was executed on an RNA-Seq dataset derived from HIV-1-infected primary T lymphocytes, aiming to identify how Vpr presence can influence the transcriptomic footprint left by HIV-1 on its host. Finally, Vpr’s ability to hijack and redirect its host’s proteasomal activity was studied in the context of two putative protein targets previously identified through non-targeted proteomics: TCF7 and G3BP1. The present work demonstrated that the role virion-delivered Vpr plays in supporting the establishment of a productive HIV-1 infection is highly reliant on its ability to induce the activation of NF-AT, as artificially inhibiting this transcriptional factor completely curtailed Vpr’s characteristic boost in viral productivity and spread. The aforementioned bioinformatic analyses revealed that Vpr-mediated NF-AT induction leads to the transcriptional reprogramming of the host T cell, differentially affecting a variety of physiological processes, including cell cycle progression, ribosome assembly, protein translation, immune & inflammatory function, intracellular signaling, and cell proliferation, amongst others. In addition, this study established the mechanism whereby Vpr leads to the proteasomal degradation of TCF7, a trans-acting factor of central relevance towards T cell development, differentiation, and survival. Taken together, these results establish Vpr-mediated NF-AT activation as a central mechanism through which HIV-1 reprograms T cell physiology to enhance viral replication, expanding Vpr’s array of virus-supporting roles and illustrating their eventual outcome on T cell physiology. Future work ought to prioritize validating many of these phenomena in primary CD4+ T cells, in parallel exploring the downstream effects of Vpr-targeted protein degradation on T cell differentiation, exhaustion, as well as in the establishment and reactivation of potential HIV-1 reservoir populations.

2026-06-09T00:00:00Z Werner, Julia Helga http://hdl.handle.net/10900/180431 2026-06-09T01:00:15Z 2026-06-08T00:00:00Z

Hardware-Aware Machine Learning Methods for Medical Edge Devices Werner, Julia Helga Real-time embedded edge devices are of great importance in many applications, such as autonomous driving, robots or smartwatches. Additionally, various medical procedures involve embedded, small in-body sensor edge devices. Equipping such devices with artificial intelligence can improve the procedures by incorporating new functionalities. However, this is often accompanied by a high demand for energy and computational resources if the models are not optimized accordingly. For some applications involving in-body edge devices equipped with machine learning models, the neural network parameters must be stored directly on-device and the model executed locally. Notably, resource-constrained devices impose stringent requirements on machine learning models in respect to on-chip area and electrical energy consumption. These restrictions need to be considered in the final model design. Deep learning methods involving neural networks with millions or even billions of parameters or operations that cannot simply be transferred to hardware, are not a viable solution. Furthermore, there is a necessity of using lightweight, quantized models in fixed-point representation to realize efficient inference on hardware. Storing model parameters in lower precision potentially impairs the overall performance of the classifier, which needs to be addressed by dedicated techniques, such as hardware-aware training. Additionally, potential challenges involve general data sparsity and class imbalances, which often occur in medical datasets since pathologies are naturally underrepresented compared to healthy samples. Importantly, if well-designed, machine-learning-based decision models provide new energy-saving functionalities that can lower the energy demand of the whole system. This thesis specifically addresses these problems by examining two important medical applications: the Video Capsule Endoscopy, a methodology to investigate the otherwise inaccessible small intestine using a small, pill-sized capsule and seizure detection using neuroimplants intended for drug-resistant epilepsy patients. The main objective of this thesis is to overcome the described challenges and design artificial intelligence-based classification models suitable for tiny edge devices as present in both introduced medical applications. It is further expected that other medical applications can benefit from the presented methods as well. Overall, this work is dedicated to the development of hardware-aware, specialized machine learning techniques for the Video Capsule Endoscopy and preictal seizure detection. The approaches are tailored for an on-device application, providing the groundwork for future innovations and enhancements, such as an actively controlled capsule. For both applications, hybrid models are proposed, combining machine learning classifiers based on deep neural networks with time-series techniques, such as Hidden Markov Models, to solve these challenges. The resulting methods are accurate, highly efficient and are verified on FPGA-based hardware demonstrators to measure their power consumption. This enhances both medical procedures involving low-power edge devices without increasing the energy demand of the whole system.

2026-06-08T00:00:00Z Eck, Jessica http://hdl.handle.net/10900/180430 2026-06-09T01:02:36Z 2026-06-08T00:00:00Z

Indirect Galactic Dark Matter Search with JUNO: A Machine Learning-Enhanced Sensitivity Study from MeV to GeV Eck, Jessica Die Natur der Dunklen Materie (DM), die ungefähr 26% des Energieinhalts des Universums ausmacht, bleibt eine der zentralen offenen Fragen der modernen Physik. Diese Arbeit widmet sich der indirekten Suche nach Dunkler Materie durch monoenergetische Neutrinos mit dem Jiangmen Underground Neutrino Observatory (JUNO)-Detektor, die den gesamten Massenbereich von mχ = 15 MeV bis 10 GeV abdeckt. Die Analyse geht von DM-Selbstannihilation via χχ → νℓν̄ℓ in der Milchstraße mit einer demokratischen Neutrino-Flavor-Komposition aus. Das Primärziel dieser Arbeit ist es, die Ausschlusssensitivität von JUNO auf den thermisch gemittelten Annihilationsquerschnitt unter Verwendung eines Bayes'schen Rahmens mit Markov-Chain-Monte-Carlo (MCMC)-Sampling zu bestimmen. Systematische Unsicherheiten werden durch log-normale Störparameter berücksichtigt, und die Robustheit gegen statistische Schwankungen in realen Messungen wird durch Toy-Monte-Carlo-Studien quantifiziert. Um diese Analyse durchzuführen, sind präzise Asimov-ähnliche Vorhersagen der Signal- und Hintergrundspektren essentiell. Daher werden unter Verwendung umfangreicher Monte-Carlo-Simulationen basierend auf GENIE und der vollständigen JUNO-Detektor-Simulation Signal- und Hintergrundspektren im gesamten Energiebereich modelliert. Der niederenergetische diffuse Supernova-Neutrino-Hintergrund (DSNB) wird unter Berücksichtigung verschiedener theoretischer Modelle einbezogen, während die atmosphärischen Neutrinoflüsse zum JUNO-Standort unter Annahme von Drei-Flavor-Oszillationen mit Materie-Effekten propagiert werden. Energieabhängige Selektionsstrategien werden entwickelt, um das Signal-zu-Hintergrund-Verhältnis in jedem Regime zu optimieren. Im MeV-Regime ermöglicht eine neue auf maschinellem Lernen (ML) basierte Vertex-Rekonstruktion mit einer Auflösung von ungefähr 18 cm eine Topologie-basierte Selektion für den inversen Betazerfall (IBD) kombiniert mit einer anschließenden Pulsformdiskriminierung (PSD). Im sub-GeV-Regime werden eine Flavor-basierte Selektion unter Verwendung von ML-basierter Teilchenidentifikation (PID) sowie eine topologische Null-Neutronen-Selektion eingeführt, um charakteristische Spektralmerkmale einer monoenergetischen Neutrinoquelle zu verstärken. Im GeV-Regime unterdrückt die Richtungsselektion um das Galaktische Zentrum (GC) den nahezu isotropen atmosphärischen Hintergrund, während ein großer Anteil des DM-Signals erhalten bleibt. JUNO kann die bestehenden Super-Kamiokande (SK)-Grenzen in dem Massenbereich mχ ≈ 15 MeV bis 1 GeV nach 10 Jahren Datennahme um etwa eine Größenordnung verbessern, wobei statistische Unsicherheit systematische Effekte dominiert. Für Einjahres-Szenarien ist die Ausschlusssensitivität im MeV-Regime konkurrenzfähig mit aktuellen Grenzen, während der sub-GeV-Bereich ein 5σ-Entdeckungspotenzial für Massen oberhalb von ungefähr 0,2 GeV erreicht. Im GeV-Regime sind Verbesserungen aufgrund der Winkelauflösung begrenzt, aber eine konkurrenzfähige, unabhängige Überprüfung bestehender Grenzen wird erreicht. Die Analyse wird auf p-Wellen-Annihilation durch J-Faktor-Reskalierung erweitert, und die in dieser Arbeit bestimmten modellunabhängigen Flussgrenzen stellen ein allgemeines Ergebnis dar, das auf jede Quelle monoenergetischer Neutrinosignale anwendbar ist.; The nature of dark matter (DM), which constitutes approximately 26% of the energy content of the universe, remains one of the central open questions in modern physics. This work is dedicated to the indirect search for dark matter via monoenergetic neutrinos with the Jiangmen Underground Neutrino Observatory (JUNO) detector, covering the entire mass range from mχ = 15 MeV to 10 GeV. The analysis assumes dark matter self-annihilation via χχ → νℓν̄ℓ in the Milky Way, with a democratic neutrino flavor composition. The primary goal of this work is to determine the exclusion sensitivity of JUNO on the thermally averaged annihilation cross section using a Bayesian framework with Markov chain Monte Carlo (MCMC) sampling. Systematic uncertainties are accounted for through log-normal nuisance parameters, and the robustness against statistical fluctuations in real measurements is quantified through toy Monte Carlo studies. To perform this analysis, precise Asimov-like predictions of signal and background spectra are essential. Therefore, using extensive Monte Carlo simulations based on GENIE and the full JUNO detector simulation, signal and background spectra are modeled across the entire energy range. The low-energy diffuse supernova neutrino background (DSNB) is incorporated considering different theoretical models, while atmospheric neutrino fluxes are propagated to the JUNO site assuming three-flavor oscillations including matter effects. Energy-dependent selection strategies are developed to optimize the signal-to-background ratio in each regime. In the MeV regime, a new machine-learning-based vertex reconstruction achieving approximately 18 cm resolution enables a topology-based selection of inverse beta decay (IBD) combined with a subsequent pulse-shape discrimination (PSD). In the sub-GeV regime, a flavor-based selection using machine-learning-based particle identification (PID) as well as a topological zero-neutron selection are introduced to enhance characteristic spectral features of a monoenergetic neutrino source. In the GeV regime, directional selection around the Galactic Center suppresses the nearly isotropic atmospheric background while retaining a large fraction of the dark matter signal. JUNO can improve existing Super-Kamiokande (SK) limits by approximately one order of magnitude in the mass range mχ ≈ 15 MeV to 1 GeV after 10 years of data taking, with statistical uncertainty dominating systematic effects. For one-year scenarios, exclusion sensitivity is competitive with current limits in the MeV regime, while the sub-GeV range achieves 5σ discovery potential for masses above approximately 0.2 GeV. In the GeV regime, improvements are limited by angular resolution, but a competitive, independent verification of existing limits is achieved. The analysis extends to p-wave annihilation through J-factor rescaling, and the model-independent flux limits determined in this work represent a general result applicable to any source of monoenergetic neutrino signals.

2026-06-08T00:00:00Z