134125 research outputs found
Sort by
SpectralKAN: Weighted Activation Distribution Kolmogorov–Arnold Network for Hyperspectral Image Change Detection
No full textKolmogorov-Arnold networks (KANs) represent data features by learning the activation functions and demonstrate superior accuracy with fewer parameters, FLOPs, GPU memory usage (Memory), shorter training time (TraT), and testing time (TesT) when handling low-dimensional data. However, when applied to high-dimensional data, which contains significant redundant information, the current activation mechanism of KANs leads to unnecessary computations, thereby reducing computational efficiency. KANs require reshaping high-dimensional data into a one-dimensional tensor as input, which inevitably results in the loss of dimensional information. To address these limitations, we propose weighted activation distribution KANs (WKANs), which reduce the frequency of activations per node and distribute node information into different output nodes through weights to avoid extracting redundant information. Furthermore, we introduce a multilevel tensor splitting framework (MTSF), which decomposes high-dimensional data to extract features from each dimension independently and leverages tensor-parallel computation to significantly improve the computational efficiency of WKANs on high-dimensional data. In this paper, we design SpectralKAN for hyperspectral image change detection using the proposed MTSF. SpectralKAN demonstrates outstanding performance across five datasets, achieving an overall accuracy (OA) of 0.9801 and a Kappa coefficient (K) of 0.9514 on the Farmland dataset, with only 8 k parameters, 0.07 M FLOPs, 911 MB Memory, 13.26 s TraT, and 2.52 s TesT, underscoring its superior accuracy-efficiency trade-off. The source code is publicly available at https://github.com/yanhengwang-heu/SpectralKAN.No Full Tex
Fluent: Round-efficient secure aggregation for private federated learning
No full textFederated learning (FL) facilitates collaborative training of machine learning models among a large number of clients while safeguarding the privacy of their local datasets. However, FL remains susceptible to vulnerabilities such as privacy inference and inversion attacks. Single-server secure aggregation schemes were proposed to address these threats. Nonetheless, they encounter practical constraints due to their round and communication complexities. This work introduces Fluent, a round-and communication-efficient secure aggregation scheme for private FL. Fluent offers improvements compared to state-of-the-art solutions, i.e., BBGLR (Bell et al., 2020) and Flamingo (Ma et al., 2023): (1) it eliminates the need for iterative handshakes and secret sharing operations by efficiently reusing the shares across multiple training iterations, while maintaining an equivalent level of security assurance as BBGLR and Flamingo; (2) it accomplishes both the consistency check and weight unmasking in one round, thereby reducing the communication complexity. With these constructions, Fluent achieves the fewest communication rounds (i.e., two in the collection phase) in the malicious server setting, in contrast to at least three rounds in BBGLR and Flamingo. This significantly reduces the latency for geographically distributed clients; (3) In addition, a dynamic variant with a participant selection algorithm and multilevel (hierarchical) secret sharing scheme was introduced. It facilitates dynamic client joining, hence enhancing the flexibility and scalability of Fluent. We implemented Fluent and compared it with BBGLR and Flamingo. Experimental results based on the EMNIST dataset show that Fluent achieves significant enhancements in training efficiency. Specifically, Fluent achieves improvements of 6.8× and 2.4× in training time against BBGLR and Flamingo respectively, without compromising accuracy.No Full Tex
A consensus of international experts on definition, sampling, treatment, and prevention of peripheral extracorporeal membrane oxygenation cannula-site infection obtained by the Delphi method: the SAVECMO study
No full textBackground: Nosocomial infections are common in patients receiving extracorporeal membrane oxygenation (ECMO), with ECMO cannula-site infections (ECMO-CSI) being the most frequent infections directly related to the ECMO run. These infections can significantly impact patient outcomes. Currently, no adult guidelines exist for the prevention, diagnosis, and/or treatment of peripheral ECMO-CSI, resulting in heterogeneity in both clinical practice and research findings. Methods: We conducted a Delphi study involving 39 international experts in ECMO management. The experts participated in four Delphi rounds to reach consensus on various aspects of ECMO-CSI complicating peripheral ECMO (central ECMO excluded), including definition, clinical suspicion, diagnostic methods, preventive measures, and treatment. Consensus was defined as ≥ 70% agreement among experts on each proposed item. Results: The Delphi process established consensus on key aspects of ECMO-CSI. Experts agreed on clinical scenarios that warrant suspicion of ECMO-CSI, such as purulent discharge and local inflammatory signs. Standardized sampling techniques, including swabs and purulent drainage aspiration, were recommended, while others were rejected. Definitions were clarified, specifying that ECMO-CSI is defined by the isolation of a pathogen through local microbiological sampling and the presence of purulent discharge or local inflammatory signs. Among the preventive measures, the use of chlorhexidine-impregnated or semipermeable polyurethane dressings, unchanged for 7 days unless soiled or bleeding, was recommended, whereas systematic antibiotic prophylaxis, even for surgical ECMO, was not recommended. Conclusion: This study presents an international expert consensus focusing on peripheral ECMO-CSI, providing a standardized framework to improve clinical management and facilitate future research. The consensus aims to enhance patient outcomes and support evidence-based guidelines in this complex field.No Full Tex
Coupled climate–land-use interactions modulate projected heatwave intensification across Africa
Full text linkAfrica faces intensifying heatwave hazards with far-reaching impacts on health, agriculture, and economic stability. Here, we assess changes in heatwave attributes, including duration, frequency, timing, and intensity, across nine African regions using 10 global numerical climate model simulations, with a multivariate bias-correction method applied to the model output. Under a high-end emissions scenario (SSP585), an upper limit projection, Western South Africa is projected to experience more than a 12-fold increase in heatwave duration and frequency. We apply Shapley Additive Explanations, an explainable artificial intelligence technique, to quantify how environmental drivers shape these changes, revealing strong regional contrasts arising from interactions among temperature, humidity, and land-surface modification. Temperature and humidity together account for more than 35% of projected increases in several regions, amplified by cropland and pasture expansion. A moderate mitigation scenario (SSP370) substantially reduces heatwave occurrence in West Africa, offering actionable guidance for land management and early-warning systems.Full Tex
Adolescent Pregnancy Among Girls Known to Child Protection Services: An Australian Longitudinal Cohort Study Using Data Linkage
No full textIn a population cohort of 44,216 adolescent girls, about 73.5% of those who became pregnant were known to child protection services. Exposure to higher levels of child protection response was associated with increased cumulative incidence and an earlier age of pregnancy.No Full Tex
Frequency and C:N:P stoichiometry of organic inputs determines intensity of net C balance in paddy soils
No full textThe frequency and nutrient composition of organic inputs jointly regulate soil organic carbon (SOC) dynamics, but their interactive effects on microbial carbon use efficiency (CUE), priming effects (PE), and net soil C balance remain poorly understood in flooded paddy systems. We performed a 40-day incubation experiment using a 13C-labeled simulated root-exudate mixture (glucose:oxalic acid:alanine, 65:30:5) under two input modes (single substrate input vs. continuous substrate input) and four C:N:P stoichiometries. Single substrate inputs generated an early pulse in labile-C mineralization that was 39–64 % greater than under continuous addition, and mineralization rates declined with increasing nutrient supply. Single-pulse addition triggered an early peak in the metabolic quotient (qCO2) and lower tracer-based CUE, whereas continuous addition maintained steadier microbial activity and higher CUE. The C:N:P stoichiometry of the added substrate strongly controlled C partitioning: stoichiometrically balanced inputs reduced CO2–C losses and increased 13C incorporation into microbial biomass and SOC pool. Pulse inputs typically induced negative PEs, whereas continuous inputs tended to cause positive PEs. Therefore, the net C balance was consistently greater following single substrate inputs than following continuous inputs; across nutrient treatments, single pulses produced substantially larger short-term C retention. Combining 13C tracing, enzyme assays and kinetic modelling, we demonstrate that under balanced nutrient inputs, microbes respire less of the added C and allocate more into biomass and necromass, which subsequently contributes to more stable SOC pool. This study provides mechanistic guidance for using C:N:P-balanced amendments to increase SOC retention in flooded cropping systems.No Full Tex
Global, regional, and national burden of chronic respiratory diseases and impact of the COVID-19 pandemic, 1990–2023: a Global Burden of Disease study
No full textChronic respiratory diseases, including chronic obstructive pulmonary disease (COPD), asthma, pneumoconiosis, interstitial lung disease (ILD) and pulmonary sarcoidosis, are major global causes of mortality and morbidity. Although the COVID-19 pandemic has influenced acute respiratory health, its impact on chronic respiratory conditions remains unclear. We estimated the global, regional and national burden of chronic respiratory diseases from 1990 to 2023, including risk factors, and evaluated how these burdens have shifted during the COVID-19 pandemic using the Global Burden of Disease Study 2023. In 2023, chronic respiratory diseases accounted for 569.2 million (95% uncertainty interval (UI), 508.8–639.8) cases and 4.2 million (3.6–5.1) deaths. The age-standardized death rate declined by 25.7% globally from 1990 to 2023 despite an increase in ILD and pulmonary sarcoidosis. Mortality declined in younger males, especially for asthma, whereas older adults experienced a rise in ILD and pulmonary sarcoidosis. Smoking was the primary risk factor for COPD, whereas high body mass index and silica exposure were key risk factors for asthma and pneumoconiosis. During the pandemic, the incidence of chronic respiratory diseases increased modestly, but the decline in mortality rates became more pronounced, highlighting the need for sustained global attention and action to address their long-term burden.No Full Tex
Engraved Limestone Block from Švédův stůl Cave, Czech Republic
No full textWhile research into Magdalenian lifeways and artistic traditions continues strongly in the ‘classic heartland’ centred on France and northern Spain, excavations at Palaeolithic localities in eastern Central Europe are also uncovering new evidence for this archaeological culture. Here we present a block of limestone which features two engraved surfaces, including at least one clear depiction of a horse. Recovered from the Moravian site of Švédův stůl Cave, this artefact may have originally been part of a parietal artwork which then detached and was incorporated into the cultural deposit, or it was engraved after the palm-size block had spalled from the shelter wall. We describe this piece, rare for this region and time period, and discuss its connections with the better understood and much richer corpus of Magdalenian art to the west.No Full Tex
Experimental investigation on liquefaction resistance of saturated silty soil: effects of clay content and intergranular void ratio
No full textSoil liquefaction remains a significant threat to infrastructure built on saturated silty deposits. To evaluate its key influencing factors, this study investigates the effects of clay content (CC), fines content (FC) and intergranular void ratio (ei) on the liquefaction resistance of saturated silty soil. A series of undrained cyclic triaxial tests were conducted on reconstituted specimens with varying CC and FC. The results indicate that the cyclic resistance ratio (CRR) does not correlate well with fines or clay content alone. To quantitatively assess liquefaction resistance, the concept of the intergranular void ratio of silty soil is introduced, which better reflects the active soil skeleton state from a microstructural perspective by considering particle contact mechanisms. Analysis of the experimental data reveals a negative power-law relationship between ei and CRR. For specimens with CC ≤ 8%, this relationship is quantified as CRR = 0.362·\:{e}_{i}^{-0.771}with R2 = 0.94. This study establishes ei as a superior state parameter and provides a new perspective on evaluating the liquefaction potential of silty soil by linking microstructural parameters to macro-scale behavior.No Full Tex
Engineering the Local Electronic Microenvironment via Interfacial Chelation for Efficient CO2 Photoreduction Toward CH4
No full textThe photocatalytic conversion of CO2 into hydrocarbons using sustainable solar energy offers a promising strategy to address the global energy crisis and achieve carbon neutrality. However, conventional p-block photocatalysts are often limited by inefficient electron transfer, which restricts the reaction to a two-electron reduction pathway, primarily yielding CO and impeding the formation of high-value hydrocarbons like CH4. Herein, we construct a novel BiOCl-BiO(HCOO) heterostructure (denoted as BiOCH), which features interfacial chelating interactions between the [Bi2O2]2 + and [HCOO]- layers within the BiO(HCOO) component, for efficient photocatalytic CO2 reduction to CH4. This unique heterostructure broadens the light absorption spectrum and facilitates the separation of photoinduced charges. More importantly, the interfacial Bi─O chelation in BiO(HCOO) modulates the local electronic microenvironment of Bi sites. Mechanistic studies reveal that this modulation enhances the coupling between the C-2p orbital of the *CHO intermediate and the Bi-p orbital, thereby lowering the Gibbs free energy barrier for the critical *CO-to-*CHO step and promoting CH4 generation. Consequently, the optimized BiOCH catalyst achieves a remarkable CH4 production rate of 42.95 µmol·g- 1·h- 1 with a high electron selectivity of 95.38%. This work provides a novel design strategy of organic-inorganic hybrid layered structures for steering photocatalytic CO2 reduction toward value-added hydrocarbons.No Full Tex