143174 research outputs found
Sort by
Delays, cooperation and pairwise comparisons in sequential decision-making processes
No full textThis thesis presents novel algorithms for sequential decision-making in uncertain environments, focussing on the challenges introduced by complex feedback structures. Specifically, we study scenarios that frequently arise in real-world applications, including delayed feedback, cooperation in multi-agent systems, and learning from pairwise comparisons.
First, we study delayed feedback. Here, the decision-maker sequentially selects actions, but the corresponding feedback arrives later. For generalised linear bandits, we propose an algorithm and prove an upper bound on its regret. This guarantee shows that the delays cause a small additive penalty on top of what is possible with immediate feedback, substantially improving upon existing results. Subsequently, we extend the study of delayed feedback to episodic reinforcement learning and propose two approaches to handle the delays. For each approach, we provide regret bounds for numerous immediate feedback algorithms and prove that the price for delays is an additive penalty.
Second, we study problem settings where numerous cooperative decision-makers interact with the environment. For these problems, we propose a black-box algorithm that takes any single-agent algorithm as input and turns it into a cooperative multi-agent algorithm. Additionally, we prove that the black-box algorithm transfers the theoretical guarantees of any single-agent algorithm to the multi-agent setting. These guarantees are comparable to, or better than, existing works.
Finally, we study learning from pairwise comparisons, where the decision-maker selects two actions and receives the preferred action as their feedback. For this setting, we propose an algorithm that collects batches of preferences and uses this information to learn a policy that generates preferable actions. We prove that the cumulative sub-optimality of our algorithm grows sub-linearly without requiring strong assumptions on the preference structure.
Overall, this thesis provides practical algorithms with theoretical guarantees. These results make a step towards bridging the gap between idealised models for sequential decision-making and real-world challenges.Open Acces
CRISPR typing and phage content of colonizing Group B Streptococci from healthy Egyptian women
No full textBackground: Streptococcus agalactiae or Group B Streptococcus (GBS) causes serious infections in neonates with a particularly high burden of disease in Africa. Maternal vaginal colonization is the primary source of neonatal transmission. Molecular surveillance of the maternal GBS population is crucial for informing maternal vaccine development and monitoring of the global circulation of GBS clones.
Methods: The current study analyzes the structure and diversity of the clustered regularly interspaced palindromic repeat (CRISPR)-associated (Cas) system and phage content in colonizing GBS isolates collected from healthy pregnant women from Ismailia, Egypt. The isolates were characterized by whole-genome sequencing within the global JUNO project.
Results: CRISPR arrays and phages were detected in a vast majority of GBS isolates. A strong congruence was observed between multilocus sequence typing (MLST), CRISPR profile, and phage content. Region-specific sequence types (STs) observed only in Africa were distinguishable from other lineages.
Conclusions: CRISPR typing is a promising low-cost tool for investigating the population structure of GBS clones, particularly in middle- and low-income countries
Online identity inference, impersonation and mitigation strategies
No full textOnline identities in social platforms and generative models are increasingly at risk as adversaries exploit unique identifiers for malicious ends. This thesis examines two complementary perspectives on such vulnerabilities: squatted usernames on social networks and attribute leakage in text-guided diffusion (T2I) models.
First, we systematically investigate how attackers capitalize on near-duplicate username registrations on X, confusing users and enabling large-scale impersonation. To illuminate this phenomenon, we develop the username-generation tool UsernameCrazy, inspecting hundreds of thousands of variants derived from celebrity accounts. Our measurements reveal a concerning ecosystem of suspended and active squatted usernames, many featuring profile pictures and names closely matching the originals to bolster deception. We further demonstrate how user mismentions and X’s search algorithm magnify these threats. To mitigate them, we propose SQUAD, a framework integrating UsernameCrazy, achieving a 94% F1-score in detecting suspicious squatted accounts.
While username squatting targets explicit digital identities, online personas are increasingly shaped by more subtle signals—particularly in the era of generative AI. We extend our analysis to emerging T2I models and their potential to leak sensitive traits-such as authorship or dementia status-through seemingly benign image outputs. By constructing adversarial pipelines that leverage image augmentation and text–image embedding models, we achieve up to 0.877% Top-5 accuracy in attributing images across 100 authors, and 0.75% accuracy in inferring dementia status (using the ADReSS dataset). These inferences are robust against diverse training sets, independent of classifier choice, and remain resilient to standard mitigation strategies.
Together, these findings reveal how adversaries can hijack identifiers-whether user-chosen names on a social platform or latent attributes embedded in generative images-to threaten security and privacy. By characterizing these overlooked risks and introducing detection and protection mechanisms, this thesis advances our collective understanding of modern identity vulnerabilities and highlights avenues to safeguard online communities.Open Acces
One-pot sol–gel synthesis of Sr/Ca-doped silica nanoparticles for osteogenic therapy in osteoporosis
No full textOsteoporosis affects more than 200 million people worldwide, with an osteoporotic fracture occurring approximately every 3 seconds; with ageing populations, its prevalence continues to rise, yet it remains under-diagnosed and under-treated. Strontium- and calcium-doped mesoporous bioactive glass nanoparticles (BGNPs) are promising due to their ability to combine bioactive bone-regenerative function with controlled therapeutic ion release. We optimized a one-step sol–gel (modified Stöber) synthesis by varying the solvent system (pure water vs. 1:1 ethanol/water) to control BGNP size and morphology and assessed their effects on pre-osteoblasts (MC3T3-E1). Characterization by electron microscopy, X-ray photoelectron spectroscopy, and ^29Si MAS NMR showed that ethanol inclusion yielded smaller, uniform spherical particles (74 ± 5 nm), whereas water alone produced significantly larger particles (224 ± 42 nm). Both Sr2+ and Ca2+ were incorporated as network modifiers within an amorphous silicate framework, with no crystalline phases. Cytocompatibility assays revealed a size-dependent response: larger particles reduced cell viability at 1 μg/mL, while both sizes were biocompatible at 0.1 μg/mL. At the non-toxic concentration of 0.1 μg/mL, BGNPs enhanced alkaline phosphatase activity, promoted osteogenic differentiation, and exhibited antioxidant activity by scavenging tert-butyl hydroperoxide-induced free radicals. These results indicate that solvent-controlled synthesis effectively tunes BGNP size without disrupting silicate network integrity, and that properly sized Sr/Ca-doped BGNPs support both osteogenic and antioxidant responses, making them strong candidates for advanced therapeutic approaches in osteoporosis treatment
A novel biplex Onchocerca volvulus rapid diagnostic test evaluated among 3- to 9-year-old children in Saridi, South Sudan
No full textBackground: Point-of-care diagnostic tests are essential for confirming Onchocerca volvulus transmission in remote, resource-limited, onchocerciasis-endemic communities. In Maridi, South Sudan, we field-tested a novel “biplex A” rapid diagnostic test (RDT) developed by Drugs & Diagnostics for Tropical Diseases (DDTD), San Diego, California. Methods: In February 2023, children aged 3–9 years were recruited from study sites at different distances from the Maridi Dam, a known blackfly breeding site. O. volvulus antibodies were detected using the DDTD biplex A RDT, which detects antibodies to Ov16 and OvOC3261 at test line 1 and to Ov33.3 and OvOC10469 at test line 2, along with the commercially available Ov16 SD Bioline RDT. Both tests were performed on whole blood obtained via finger prick. The feasibility and acceptability of the DDTD biplex A RDT were assessed, and its results were compared with those of the Ov16 SD Bioline RDT. Results: A total of 239 children participated in the study. The anti-Ov16 seroprevalence detected by the Ov16 SD Bioline RDT was 30.2% (72/239), with the highest prevalence observed in children living closest to the Maridi Dam (p < 0.001). Testing with the DDTD biplex A RDT was determined to be feasible, acceptable, and easy to use in a field setting. The DDTD biplex A RDT test line 1 (anti-Ov16 and anti-OvOC3261) was positive in 35.1% (84/239) of children, while test line 2 (anti-Ov33.3 and anti-OvOC10469) was positive in 18.4% (44/239). Both lines were simultaneously visible in 15.5% (37/239). Conclusions: The DDTD biplex A RDT prototype was user-friendly and practical for field deployment. However, additional research is needed to evaluate its performance relative to the commercially available Ov16 SD Bioline RDT. The high anti-Ov16 seroprevalence that was observed underscores the ongoing O. volvulus transmission near the Maridi Dam. Strengthening the onchocerciasis elimination program in Maridi should be prioritized as a critical public health intervention
The importance of national and disease-specific characteristics to estimate disease burden in the GBD: a comparison for haemoglobin disorders in Italy.
No full textScaling out community conservation initiatives: experts identify economic and social benefits, compatibility with needs, and external support as key
No full textCommunity-based natural resource management is a common strategy for conserving biodiversity, but little is known about how such initiatives can scale appropriately and widely. We interviewed 80 experts across 5 widely adopted community-based initiatives (in Chile, Nepal, Namibia, Madagascar, and Fiji) to understand their perspectives on the drivers of adoption and spread. We used general elimination methodology and diffusion of innovation theory to identify and rule out possible explanations. Factors consistently considered influential were economic and social benefits, compatibility with needs, support and facilitation by extension agencies, and the presence of international organizations. Initiatives aiming to scale should be designed to be flexible and aligned with adopters’ needs, and external organizations should coordinate resources for scaling out. Dependence on external support underscores the need for quality assistance, good practices by external actors, and understanding power relations and fairness, as well as the need to temper donor and policy expectations of scaling beyond supported and appropriate sites
High epilepsy prevalence and excess mortality in onchocerciasis-endemic counties of South Sudan: a call for integrated interventions
No full textBackground
Epilepsy is a major health concern in onchocerciasis-endemic regions with intense transmission, where the infection is associated with a high epilepsy burden. This study investigated epilepsy prevalence and mortality in five onchocerciasis-endemic counties of South Sudan, and the association between onchocerciasis transmission and epilepsy, including probable nodding syndrome (pNS).
Methodology
House-to-house cross-sectional surveys (2021–2024) identified persons with suspected epilepsy (sPWE) and retrospectively documented deaths among sPWE and individuals without epilepsy (IWE). Epilepsy diagnoses, including pNS, were confirmed by trained clinicians. Ongoing transmission was assessed using anti-Ov16 seroprevalence in children aged 3‒9 years. Age- and sex-standardised epilepsy, pNS and anti-Ov16 prevalence were calculated, along with age- and sex-standardised mortality rates and standardised mortality ratios (SMRs) with 95% confidence intervals (95%CIs), using IWE as the reference population. Weighted arcsin-transformed linear regression was used to explore the association between epilepsy and anti-Ov16 prevalence.
Principal findings
Among 34,019 individuals screened, 166 deaths occurred in 3,101 person-years for sPWE versus 466 deaths in 63,420 person-years for IWE. Epilepsy prevalence was 4.1% (range: 2.3-7.1%), and pNS prevalence was 1.5% (range: 0.6-2.2%). Anti-Ov16 seroprevalence among children was 23.3% (range: 1.4-44.1%). Each 1.0 percentage point increase in standardised anti-Ov16 seroprevalence was statistically significantly associated with an average rise of 0.10 percentage points in standardised epilepsy prevalence and 0.04 percentage points in standardised pNS prevalence. Median age at death was lower for sPWE (20 years) than IWE (38 years; Mann-Whitney U-test p-value < 0.0001). Standardised mortality rates per 1,000 person-years were statistically significantly higher in sPWE (67.6, 95%CI: 52.6-87.1) than in IWE (9.0, 95%CI: 7.8-10.3). The overall SMR was 6.9 (95%CI: 5.9-8.0), indicating sPWE were seven times more likely to die than IWE.
Significance
The high epilepsy burden in onchocerciasis-endemic areas is driven by elevated epilepsy prevalence and mortality. Integrated onchocerciasis and epilepsy programmes must be strengthened to decrease epilepsy incidence and ensure uninterrupted access to antiseizure medication
Phylogenetics and evolution of Digitaria grasses, including cereal crops fonio, raishan, and Polish millet
No full textBackground and Aims Millet crops in the grass genus Digitaria include white and black fonio (D. exilis and D. iburua), raishan (D. compacta) and Polish millet (D. sanguinalis), cultivated across West Africa, India, and Europe. Fonio and raishan crops are important to supporting food security and subsistence agricultural systems in rural communities, while D. sanguinalis is no longer cultivated. These crops are resilient to challenging climates. We aim to produce an integrated study of these crops: a phylogeny of the Digitaria genus including all four food species, to identify key crop wild relatives (CWRs); time-calibrated biogeographic analysis, to investigate the history and evolution of Digitaria; and morphological study to assess the transition between wild and domesticated species.
Methods We use the Angiosperm 353 target-enrichment sequencing approach to produce maximum likelihood and coalescent model nuclear phylogenies for 46 Digitaria species, and Bayesian methods to propose an evolutionary and biogeographic history for the genus. Morphology of wild and cultivated species is investigated for spikelets and growth habits using microscopy and SEM imaging.
Key Results Four distinct evolutionary lineages are found for the Digitaria crops, and we identify new close crop wild relatives D. fuscescens, D. atrofusca, D. setigera and D. radicosa, and D. ciliaris. South and eastern Africa is proposed as a likely origin of early Digitaria divergence, with crop lineages diverging from wild relatives around 2-6mya. Incomplete domestication traits are observed, including the loss of trichomes, but no clear change in appearance for spikelet or abscission zone morphologies.
Conclusions The knowledge produced in this study about Digitaria CWRs will be useful in improving crop traits through targeted breeding and physiological studies; and we also highlight the need for conservation of seed material through programmes working with local partners, for these important climate-tolerant indigenous cereals
The role of ethanol in lithium-mediated nitrogen reduction
No full textAlthough the Haber–Bosch process for industrial ammonia production is hailed by many as one of the most influential breakthroughs of the 20th century, its decarbonization and decentralization remain a critical challenge. One of the most promising and fast improving approaches is electrochemical nitrogen reduction mediated by lithium. However, the impact of electrolyte configuration on the formation of the solid electrolyte interphase (SEI) and its effect on selective nitrogen reduction is still elusive. In particular, the role of commonly added, supposedly sacrificial, proton donors on SEI chemistry and morphology remains a mystery. In this work, the impact of ethanol concentration in a 1 M LiNTf2 in THF electrolyte on SEI properties and nitrogen reduction is analyzed via a multipronged characterization approach. Post-mortem surface analysis via X-ray photoelectron spectroscopy shows a dependence in the relative proportion of LiF and Li2O on ethanol concentration, while depth profiling measurements via cluster source time-of-flight secondary ion mass spectrometry reveal increasing SEI electrolyte permeability at higher ethanol concentrations. Cryogenic electron microscopy measurements show a reduction in SEI thickness with increased ethanol concentration, as well as increased SEI homogeneity. Lithium metal is also observed only in the ethanol-free condition. Analysis of bulk SEI components via titration corroborates the observation of lithium metal in cryo-microscopy measurements, as well as showing an increase in bulk Li2–xOHx content with ethanol concentration. A narrow ‘Goldilocks’ region is revealed, where the SEI has just the right properties for efficient nitrogen reduction