134125 research outputs found
Sort by
Backdoor defense based on adversarial prediction proximity and contrastive knowledge distillation
No full textDeep neural networks (DNNs) have become indispensable across various fields; however, their susceptibility to backdoor attacks poses significant security risks. In this paper, we propose a backdoor defense scheme based on adversarial prediction proximity and contrastive knowledge distillation. This scheme not only detects poisoned models and labels but also effectively unlearns backdoors while preserving the model's benign functionality. Based on the observation that untargeted adversarial examples and poisoned samples exhibit proximity in feature space within poisoned models (i.e., adversarial prediction proximity), we first detect backdoors by analyzing changes in the prediction behavior of untargeted adversarial examples for models before and after fine-tuning. Next, we purify the poisoned model using a triplet loss that incorporates clean samples and untargeted adversarial examples. This process is guided by contrastive knowledge distillation, where a fine-tuned model acts as a “benign teacher”, and a backdoor-retained model serves as a “malicious teacher”, encouraging the poisoned model to align its feature representations with clean behavior. Comprehensive experimental results demonstrate that our scheme achieves high accuracy in detecting poisoned models and labels, even with limited access to clean samples. Furthermore, our scheme provides effective backdoor purification, while preserving the integrity and performance of models.No Full Tex
A systematic review of microplastics in coral reef ecosystems: Abundance, distribution, toxicity, and future research directions
Get PDFCoral reefs, among the most biodiverse ecosystems providing critical ecological services, are increasingly exposed to microplastics (MPs), an emerging contaminant of concern found throughout freshwater and marine environments. Despite the growing number of studies, the lack of harmonized methodologies and reporting units for MP abundance in coral ecosystems hinder meaningful comparison across studies. This inconsistency poses challenges for ecotoxicology, where coral toxicity studies rely on accurate and comparable environmental data. Here, we adopted a systematic quantitative literature review approach to analyze 125 publications including 69 studies on the abundance of MPs in coral reefs environment and 56 studies on the toxicity of MPs on corals. Various MP forms, polymer types, colors and sizes have been reported. Reported MP concentrations in coral reef environments varied among sampled locations, spanning over six orders of magnitude in water (0.046 to 389,280 MPs/m3), up to five orders of magnitude in sediment (3.35 to 431.67 MPs/kg), and over two orders in coral tissue (0.03 to 14.18 MPs/g). To date, knowledge of the impact of environmentally relevant MPs on coral health remains limited. However, studies suggest these effects can be both species-specific and depend on polymer type and MP shape. Reported impacts include reduced growth and calcification, altered photosynthetic efficiency, increased mucus production, and in some cases coral bleaching and tissue necrosis. Furthermore, this review highlights the lack of appropriate quality assurance and quality control (QA/QC) protocols and discusses recommendations for future studies to enhance the data quality and relevance.Full Tex
Distance Learning-Based Prototypical Network With Multi-Domain Adaptation for Few-Shot Hyperspectral Medical Image Classification
Get PDFHyperspectral imaging (HSI) holds immense potential for medical diagnostics by capturing tissue-specific spectral signatures that facilitate precise disease detection. However, effective HSI classification in clinical settings is hindered by two main challenges: (i) the severe lack of labelled medical HSI samples constrains model training. Prototypical networks, as a few-shot learning paradigm, have been adopted to address label scarcity. However, current Euclidean-based prototypical methods typically assume equal feature variance and spherical distributions, while ignoring intraclass covariance and spectral correlations; (ii) significant domain shifts across heterogeneous medical HSI datasets undermine model generalisation, impair multi-domain interpretability, and force expensive per-dataset retraining. To overcome these limitations, we propose a novel distance-learning-based prototypical network with multi-domain adaptation for few-shot hyperspectral medical image classification. First, by embedding a class-covariance-aware Mahalanobis metric within the prototypical block, our module adapts similarity measures to each class's intrinsic spectral–spatial covariance and scale variations, thereby enhancing prototype robustness under severe label scarcity and significantly reducing misclassification compared with existing few-shot networks. Secondly, we introduce the domain-aware adapter block designed to address domain shift and multi-domain variability by dynamically fusing shared spectral–spatial representations with domain-specific characteristics via spectral integration and switchable adapters. We undertook extensive experiments on three publicly available hyperspectral medical datasets: skin dermoscopy, multidimensional choledochal, and in-vivo brain dataset. Compared to state-of-the-art classifiers, the proposed method achieved excellent performance on all three datasets, paving the way for generalisable HSI solutions in clinical workflows and biomedical research.Full Tex
Rethinking social media and mental health: The role of emotion regulation difficulties
Get PDFResearch, on the whole, does not suggest that time spent on social media is associated with risks to mental health, although it is possible there are more nuances about how people use social media. Further, evidence suggests that individuals with emotion regulation difficulties may be drawn to certain social media behaviours as a means of coping with distress. The present study aimed to examine whether emotion regulation difficulties predict patterns of social media use and, in turn, symptoms of depression and anxiety. We examined four distinct types of social media use: (1) image management-based, (2) social comparison-based, (3) negative engagement-based, and (4) passive consumption-based. Sampling 548 adults aged 18–84 years (Mage = 33.16, SD = 17.37; 401 (73.2 %) female; 128 (23.2 %) male), we tested a structural equation model to examine whether the four distinct types of social media use mediated links between difficulties in emotion regulation at Time 1 and depression and anxiety symptomology at Time 2, one week later. Results suggested that, when controlling for age, difficulties in emotion regulation significantly predicted all types of social media use and symptoms of depression and anxiety over one week. Only comparison-based social media use predicted anxiety symptoms over time. The model explained 50.1 % and 52.1 % of the variance in depression and anxiety symptoms, respectively. Taken together, these findings suggest the critical importance of emotion regulation in predicting mental health. By contrast, with the exception of social comparison and anxiety, no form of social media use predicted mental health outcomes.Full Tex
River grabbing from the source: groundwater extraction and the self-perpetuating colonial practices of dispossession in Australia's Northern Territory
Get PDFIndigenous people of the Roper River in Australia's Northern Territory (NT) have strong legal rights to most of the land through which the river flows, including its bed and banks. Yet the settler-colonial state asserts control over its waters, whether flowing across or under the surface of the land. We show that the Northern Territory government has applied lessons from its colonial history of land grabbing to thwart Indigenous claims to water, including those established via the Aboriginal Water Reserve, a mechanism of settler water law. Our focus is on regulatory procedures in the Mataranka region, where groundwater extraction is occurring on a scale that puts at stake the very existence of the Roper River and its complex of sacred springs and wetlands and threatens to undermine vital socio-cultural practices. A river grabbing frame centres the river as a living, socially unifying, and hydrologically interconnected entity. It moves the focus of justice struggles from water conceived as a disembedded, divisible resource to the relations that underlie settler-colonial state domination. Here, river grabbing is materialised through rampant water hoarding, speculation and over-extraction, which are legitimised by a racialized system of water governance that entrenches and protects non-Indigenous water use. We contribute to literature on the political contingencies of dispossession by showing how the administrative exercise of state power in water planning and allocation processes perpetuates a colonial mode of extraction that denies Indigenous peoples' legitimate expectations for water rights and real power to make decisions over the region's future.Full Tex
Co-Application of Biochar and Organic Matter With Synthetic Fertilisers Improves Nitrogen Use Efficiency, Rice Yield and Benefit–Cost Ratio: A Meta-Analysis
Get PDFOptimising the efficiency of applied nitrogen (N) fertilisers is essential to sustain agricultural systems. Substantial N losses continue through leaching, volatilisation, and denitrification processes. Co-application of organic amendments and biochar alongside synthetic fertilisers is a widely practiced strategy to enhance N retention, improve soil fertility and increase crop productivity. Previous studies have focused on the specific characteristics of soil amendments and the magnitude of yield change, while N use efficiency (NUE) and economic returns remain uncertain. This meta-analysis examined the effects of synthetic fertiliser applied alone, co-applied with biochar and co-applied with organic amendments, on crop yield, plant N uptake, NUE and economic return within rice cropping systems. Synthetic fertiliser and biochar applied alone increased rice yield by 69.2% ± 30.3 and 33.4% ± 34.9, respectively (Bootstrap 95% CI), whereas yield further increased by co-applying biochar (+104.8% ± 37.5) and organic amendments (+80.2% ± 18.2) with fertiliser compared with non-fertilised control. Co-applying organic amendments (+20.9% ± 29.7) and co-applying biochar (+35.1% ± 18.3) with synthetic fertiliser increased NUE compared with fertilised control. For rice crops under low N application (< 150 kg ha−1), co-applying biochar with fertiliser increased yields more than co-applying organic amendments (+70.1% ± 0.7 vs. +52.5% ± 0.3, respectively). Within acidic soils, co-applying biochar with fertiliser (+72.9% ± 0.4) led to higher yield than co-applying organic matter (+36.0% ± 0.9), and among soils with high organic carbon concentration, co-applying biochar with fertiliser increased yield by 97.6% ± 1.6, compared with yield increases observed by co-applying organic matter with fertiliser at 29.4% ± 0.5 and fertiliser alone at 25.6% ± 0.2. The main factors driving rice yield were N application rate, co-application method and soil organic carbon concentration. Co-applying either biochar or organic amendments did not significantly differ in benefit–cost ratio with benefit–cost ratios of 35.1% ± 9.2 and 18.1 ± 26.5, respectively compared with fertilised control. Co-applying either biochar or organic amendments with synthetic fertilisers decreased N inputs and increased economic return, therefore improving sustainability in rice cropping systems.Full Tex
Delays to Post-Operative Radiotherapy in Patients With Head and Neck Squamous Cell Carcinoma: A Mixed-Methods Analysis of Health Worker Perspectives
No full textIntroduction: Post-operative radiation therapy (PORT) delays are associated with reduced survival and higher recurrence rates in patients with head and neck squamous cell carcinoma (HNSCC). This study explores health worker perspectives on barriers and solutions to timely PORT in an Australian regional centre with a large Aboriginal and Torres Strait Islander population. Methods: This paper incorporated an explanatory sequential mixed methods design. Head and neck multidisciplinary team (HNMDT) members involved in the care of patients with HNSCC at a regional tertiary centre were invited to participate. Participants completed an online survey and underwent semi-structured focus groups/interviews. Results: Nineteen healthcare workers participated in the surveys. The three factors most attributed to PORT delays were rurality, surgery at an external facility and Indigenous status. Ten healthcare workers subsequently participated in focus groups/interviews. Three themes regarding solutions to PORT delays were identified: addressing care fragmentation, improving care coordination and ensuring cultural safety in Aboriginal and Torres Strait Islander care. Strategies proposed to address care fragmentation included streamlining referral pathways, utilising electronic reminders for clinicians and coordinators to prompt follow-up, and increasing local service provision. Solutions suggested to improve care coordination included flagging patients early for follow-up, timely organisation of transport and accommodation, and employing head and neck-specific cancer care coordinators. Strategies proposed to ensure cultural safety included instituting cultural immersion programmes, promoting earlier Indigenous Liaison Officer engagement, supporting family involvement, and developing culturally appropriate patient resources. Conclusion: This study highlights barriers and solutions to timely PORT in an Australian regional centre with a large Aboriginal and Torres Strait Islander population. Strategies aimed at the patient and systems level and formed in partnership with Indigenous peoples must be instituted to address delays.No Full Tex
Trait-explicit approaches cast new light on fragmentation’s effects on biodiversity
No full textWhile negative effects of landscape fragmentation on biodiversity are well documented, recent evidence suggests positive or neutral effects may also be common. The mechanisms driving these contrasting outcomes cannot be understood without assessing how species traits influence responses to landscape and patch characteristics. Here, we show that three key elements can enhance predictability of fragmentation effects across scales: (i) the trait distribution of the regional species pool; (ii) the relationship between taxonomic and trait diversity; and (iii) the effects of the landscape matrix on the distribution of species traits. Considering these elements will facilitate the development of generalizable hypotheses on the consequences of fragmentation across diverse taxonomic groups and regions, with broad applicability to ecology and conservation.No Full Tex
A stable environmental niche for humans in the southern Levant 70–40 ka
No full textEnvironmental drivers were likely key to human dispersals from Africa into and throughout Eurasia, but the effect of such drivers on human biogeography has yet to be resolved at high-resolution on a regional scale. Here, we probe the Levantine-Arabian region for environments favourable to human forager groups around 50 ka when a demographic wave surged across Eurasia imprinting the ancestry of all non-Africans living today. We present a set of 33 optically stimulated luminescence dates demonstrating more than 50,000-years of persistent riverine wetlands on the eastern margin of the Jordan Rift Valley at Hamra Faddan and Wadi al-Hasa—the latter hosting stratified Middle Palaeolithic artefacts indicative of frequent human presence. By reviewing and combining multiple climate proxy records, our analysis reveals permanent surplus moisture existed across much (∼70,000 km2) of the southern Levant during the interval 70–40 ka, in contrast to surrounding regions such as interior Arabia where intensified aridity and a paucity of archaeological sites primarily suggest landscape abandonment. We propose that the southern Levant offered a relatively stable, favourable environment for foraging human populations extending to the Upper Palaeolithic, during which time the region was a crucible for fostering human admixture, knowledge sharing and technological evolution. The southern Levant likely functioned as one of several population and cultural hubs in Southwest Asia during the Late Pleistocene.No Full Tex
Investigation of p-coumaric acid on intracerebroventricular lipopolysaccharide-induced spatial memory impairment and neuroinflammation in rats
No full textNeuroinflammation mediated by the activation of microglia and subsequent release of proinflammatory cytokines is a key contributor to the pathogenesis of neurodegenerative disorders. In this study, we investigated the neuroprotective effects of p-coumaric acid (PCA) in a lipopolysaccharide (LPS)-induced rat model of neuroinflammation and cognitive impairment. Neuroinflammation was induced by intracerebroventricular (ICV) administration of 150 µg/kg bacterial endotoxin LPS into the fourth ventricle of Sprague–Dawley rats, whereas PCA (160 mg/kg), Donepezil (DON, 5 mg/kg) were administered orally for a period of 14 days, post-LPS administration. PCA has been reported to be active against LPS-induced sickness behavior and chronic unpredictable mild stress models in mice, whereas DON is a centrally acting acetylcholinesterase inhibitor with documented antineuroinflammatory property. Animals were subjected to the Morris Water Maze to assess spatial memory. ICV administration of LPS caused a significant decline in cognitive ability. PCA and DON treatment effectively attenuated this LPS-induced cognitive deficits. In addition to the behavioral improvements, both treatments significantly reduced the central levels of proinflammatory cytokine, interleukin-1β, and lipid peroxidation marker, malondialdehyde levels. Our findings suggest that PCA exerts neuroprotective effects against LPS-induced neuroinflammation and cognitive impairment in rats by plausible modulation of proinflammatory cytokines and oxidative stress pathways.No Full Tex