Search Results (211)

Vision–language models (VLMs), such as Contrastive Language–Image Pretraining (CLIP), are increasingly deployed in real-world applications, including content moderation, misinformation detection, and fraud analysis, making their robustness to adversarial attacks a critical concern. While adversarial robustness has been widely studied in unimodal models, modality-specific vulnerabilities in multimodal models remain underexplored. In this work, we analyze CLIP by applying gradient-based adversarial attacks to its vision and language modalities, both independently and jointly, and evaluating performance on two multimodal classification benchmarks: the Facebook Hateful Memes dataset and a large-scale Suspicious Car Parts dataset. Using Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) attacks along with multiple adversarial retraining strategies, we show that adversarial perturbations on the image modality consistently cause the most severe and unstable performance degradation. These results demonstrate that the vision modality is the primary vulnerability in CLIP, highlighting the need for modality-specific defense strategies that focus more on the weaker modality in multimodal systems. Full article

Soil contamination in South African agriculture poses escalating threats to food security and ecosystem integrity, yet the geospatial and machine learning evidence base addressing this problem has never been systematically synthesised. This scoping review, conducted within the PRISMA-ScR framework, applied SVM-assisted screening to 2000 retrieved records, yielding a final corpus of 228 eligible studies published from 2003 to 2025. To characterise temporal, thematic, and geographic patterns in the corpus, we applied machine learning-assisted topic modelling (LDA, k = 7), logistic growth modelling, keyword co-occurrence network analysis, and technology–contaminant evidence gap matrices. Remote sensing was the dominant methodology throughout the review period (n = 142; 62.3% of studies), with machine learning rising to the highest adoption rank from approximately 2020 onwards. Logistic modelling estimated a carrying capacity of K = 292.3 (95% CI: 269–324) studies and an inflexion year of 2020.2 (95% CI: 2019.4–2021.1), projecting 90% saturation by 2028. Research effort was highly concentrated in KwaZulu-Natal and the Eastern Cape, while Pesticides/Herbicides and acid mine drainage each comprised only three corpus studies. Deep learning registered zero entries across all cells of both the technology–contaminant and technology–province evidence matrices. Targeted investment in field validation, hyperspectral and deep learning deployment for underrepresented contaminants, and interpretable modelling for regulatory defensibility are identified as priority actions for the next research cycle. Full article

This paper investigates an optimal cooperative guidance strategy for the active defense of an early-warning aircraft (EWA) escorted by two fighters against an incoming missile. The proposed framework extends classical three-body defense models (Target–Missile–Interceptor) into a more realistic four-body engagement (Target–Missile–Interceptor 1–Interceptor 2), allowing explicit coordination among multiple defenders. By projecting the 3D engagement kinematics onto two orthogonal 2D planes—a validated simplification for typical aerial combat geometries—a tractable dynamic model is obtained. Within this model, an analytical cooperative guidance law is derived using optimal control theory and the calculus of variations, minimizing a multi-objective cost function that combines miss distance, control effort, intercept geometry, and coordination terms. Extensive Monte Carlo simulations across 23 attack directions and multiple initial ranges demonstrate that the proposed method achieves an interception success rate of 99%, with an average miss distance of below 5 m. Robustness tests further confirm stable performance under target maneuver uncertainty, sensor noise, and modeling deviations. The algorithm features closed-form control commands with low computational complexity, enabling real-time onboard implementation. Full article

Chronic Kidney Disease (CKD) is a global health crisis, projected to be the fifth leading cause of death by 2040. Its progression is driven by a reinforcing loop of mitochondrial dysfunction, oxidative stress, and chronic inflammation. The AMPK-NRF2–FOXO axis serves as a central “redox-metabolic rheostat” that maintains renal homeostasis but is commonly dysfunctional in CKD. Herein, we explore the molecular crosstalk within this network, where AMPK acts as a metabolic and redox sensor, NRF2 governs the cytoprotective response, and FOXO isoforms regulate autophagy, antioxidative defense, and senescence. We highlight the functional paradoxes within the axis and evaluate the benefits and drawbacks of nutraceuticals and pharmacological agents, such as NRF2 inducer bardoxolone methyl, underscoring the necessity for context-dependent modulation. Furthermore, we examine the AMPK–NRF2–FOXO axis within the current clinical management, according to the 2024/2026 KDIGO guidelines. These guidelines reflect a shift toward a multi-targeted pharmacological approach involving metformin, SGLT2 inhibitors, GLP-1 receptor agonists, finerenone, and hypoxia-inducible factor-prolyl hydroxylase (HIF-PH) inhibitors. Full article

Scaevola taccada, as a key dominant plant in coastal ecosystems, plays an irreplaceable role in sand fixation, shoreline protection, and maintaining the ecological stability of coastal zones. To investigate the effects of drought stress on the Binghai plant Scaevola taccada seedlings, a natural drought treatment was applied. Physiological indicators were measured at 0, 10, 25, and 40 days of stress, and 5 days after rewatering. Transcriptome sequencing and long non-coding RNA (lncRNA) analysis were also conducted to reveal the drought response mechanisms and molecular regulatory networks. The results showed that: (1) Prolonged drought significantly inhibited growth, with relative height increase, leaf number, and relative water content declining by 46.8%, 37.2%, and 63.4%, respectively, at T40 compared to the control. (2) In terms of photosynthetic physiology, Rubisco activity, RCA activity, SPAD value, Fv/Fm, and qP all continuously declined with increasing stress, while NPQ increased, suggesting damage to the photosynthetic system but also the activation of energy dissipation mechanisms to alleviate photooxidative stress. (3) The antioxidant system played a crucial role in the drought response. Under drought stress, the activities of SOD, POD, and CAT, and MDA content, underwent significant changes, with antioxidant enzyme activities rebounding notably after rewatering. (4) Transcriptome analysis revealed that differentially expressed mRNAs and lncRNA-targeted genes were significantly enriched in the ‘photosynthesis’ and ‘carbon metabolism’ pathways. Key genes involved, including PSAD-1, PSAL, NPQ4, six LHCs, BAM3, BAM1, SSII-A, and FRK1, were identified as core components of the regulatory network. In summary, Scaevola taccada effectively responds to drought stress through multi-level mechanisms, including photosynthetic regulation, carbon metabolism regulation, antioxidant defense, and transcriptional reprogramming, demonstrating strong drought resistance and post-rewatering recovery potential. These findings provide scientific evidence for plant selection and application in ecological restoration projects in coastal areas in the context of global climate extremes. Full article

Cybersecurity vulnerabilities are rapidly increasing, but public understanding and awareness remain limited. Since most vulnerabilities are common, they continue to exist and to be exploited. Although there are tools, including the Open Worldwide Application Security project and the common weakness enumeration method, that provide extensive information on known security problems, their information is not structured and visually shown. The tools are ineffective in speed assessment and response. We analyzed large-scale common vulnerabilities and exposures JavaScript object notation datasets to recognize key threats, to understand the underlying cause of data breaches, and to analyze vulnerability trends. Implementing keyword gate-filling techniques and better data visualization enhances the clarity and usefulness of vulnerability information. These tools enable stakeholders to make quicker and more informed decisions and implement stronger encryption and defensive measures. Finally, the results of this study lead to broad awareness, active security, and a reactive strategy to evolving cyber threats that simplifies both governmental and average-day user recognition and response to emerging attack patterns and risks across digital platforms. Full article

For Autonomous Vehicles (AVs), recognizing traffic lights and signs is critical for safety because perception errors directly affect navigation decisions. Real-world disturbances such as glare, rain, dirt, and graffiti, as well as digital adversarial attacks, can lead to dangerous misclassifications. Current research lacks (i) temporal continuity (stable detection across consecutive frames to prevent flickering misclassifications), (ii) multi-field-of-view (FoV) sensing, and (iii) integrated defenses against both digital and natural degradation. This paper presents two principal contributions: (1) a three-layer defense framework integrating feature squeezing, inference-time temperature scaling (softmax $\tau$ = 3 without distillation training), and entropy-based anomaly detection with sequence-level temporal voting; (2) a 500 sequence dual-FoV benchmark (30k base frames, 150k with perturbations) from aiMotive, Waymo, Udacity, and Texas sources across four operational design domains. The unified defense stack achieves 79.8% mAP on a 100-sequence test set (6k base frames, 30k with perturbations), reducing attack success rate from 37.4% to 18.2% (51% reduction) and high-risk misclassifications by 32%. Cross-FoV validation and temporal voting enhance stability under lighting changes (+3.5% mAP) and occlusions (+2.7% mAP). Defense improvements (+9.5–9.6% mAP) remain consistent across native 3D (aiMotive, Waymo) and projected 2D (Udacity, Texas) annotations. Preliminary recapture experiments (n = 15 scenarios) show 2.5% synthetic–physical ASR gap (p = 0.18), though larger validation is needed. Code, models, and dataset reconstruction tools are publicly available. Full article

Drought is the most severe natural hazard threatening agricultural production. Mulberry (Morus alba L.) is an important crop for the sericulture industry, and its drought tolerance has been extensively studied. In this study, the phenotypic and physiological responses of two different mulberry tree genotypes (711 and NS8) to drought stress were investigated, with the aim of screening potential nondestructive traits and understand interrelationships. The significant reductions of digital biomass (DB), leaf area (LA), and projected leaf area (PLA) in morphological traits indicated that drought led to a decrease in mulberry yield. The change of color traits R_FarRed and R_NIR were associated with pigments and leaf morphology. Vegetation indexes were also significantly affected by drought stress. Due to their had high correlation coefficients and good linear relationships with yield, DB and LA can be used as yield proxy traits for this measure. Drought-sensitive traits were identified using PCA and correlation analysis, and the results showed that greenness (GR) was a proxy predictor of drought stress. For antioxidant defenses, CAT activity and phenolic compound content were significantly decreased. Metabolomics analysis revealed that genotype 711 exhibited 1691 differential metabolites under drought stress; these mainly comprised amino acids, lipids, and phenolic acids, which were mainly enriched in secondary metabolism and flavonoid biosynthesis. Drought also reprogrammed carbohydrate, secondary compounds, and amino acid metabolism. The results revealed that the phenotypic response of two mulberry trees to drought, as well as the integration of phenotypic traits with metabolic traits, could help us to understand drought tolerance mechanisms and benefit efficient selection and breeding of fitter genotypes. Full article

Risk management is a crucial task in project management in light of the uncertainty that is almost always inherent. This phenomenon is enhanced in projects in the defense sector, due to their high complexity and their nature as system of systems (SoS). In general, traditional risk management approaches often fall short. To address this issue, this paper explores the utilization of soft logic to improve risk management methodologies in SoS. Soft logic can address the problem of representing contradictory situations. The proposed methodology, unlike the traditional one, does not collapse the variety of estimations into a single central value, thus, it is able to perform extreme evaluations as well. An empirical analysis was conducted by means of qualitative and quantitative approaches. The findings suggest that the methodology based on soft logic may enhance decision-making processes to handle risks more effectively. Full article

Cybersecurity teams rely on signature-based scanners such as Loki, a command-line tool for scanning malware, to identify Indicators of Compromise (IOCs), malicious artifacts, and YARA-rule matches. However, the raw Loki log output delivered as CSV or plaintext is challenging to interpret without additional visualization and correlation tools. Therefore, this research discusses the creation of a web-based dashboard that displays results from the Loki scanner. The project focuses on processing and displaying information collected from Loki’s scans, which are available in log files or CSV format. DIGITRACKER was developed as a proof-of-concept (PoC) to process this data and present it in a user-friendly, visually appealing way, enabling system administrators and cybersecurity teams to monitor potential threats and vulnerabilities effectively. By leveraging modern web technologies and dynamic data visualization, the tool enhances the user experience, transforming raw scan results into a well-organized, interactive dashboard. This approach simplifies the often-complicated task of manual log analysis, making it easier to interpret output data and to support low-budget or resource-constrained cybersecurity teams by transforming raw logs into actionable insights. The project demonstrates the dashboard’s effectiveness in identifying and addressing threats, providing valuable tools for cybersecurity system administrators. Moreover, our evaluation shows that DIGITRACKER can process scan logs containing hundreds of IOC alerts within seconds and supports multiple concurrent users with minimal latency overhead. In test scenarios, the integrated Loki scans were achieved, and the end-to-end pipeline from the end of the scan to the initiation of dashboard visualization incurred an average latency of under 20 s. These results demonstrate improved threat visibility, support structured triage workflows, and enhance analysts’ task management. Overall, the system provides a practical, extensible PoC that bridges the gap between command-line scanners and operational security dashboards, with new scan results displayed on the dashboard faster than manual log analysis. By streamlining analysis and enabling near-real-time monitoring, the PoC tool DIGITRACKER empowers cyber defense initiatives and enhances overall system security. Full article

Recent projections by the Intergovernmental Panel on Climate Change indicate that global warming will turn permanent and further intensify the severity and frequency of extreme weather events (heat waves, rain, and intense droughts), with coastal regions being the most vulnerable to extreme events. Therefore, the risk of natural disasters and the associated regional impacts on water, food, energy, social, and health security represents one of the world’s greatest challenges of this century. However, conventional methodologies for monitoring these regions during extreme events are usually not available to managers and decision-makers with the necessary urgency. The aim of this study was to present a framework concept for assessing extreme flood event impacts in coastal zones using a suite of field data combined with numerical (hydrological, meteorological, and hydrodynamic) and computational (flooding) models in a virtual environment that provides a replica of a natural environment—the Patos Lagoon Digital Twin. The study case was the extreme flood event that occurred in the southernmost region of Brazil in May 2024, considered the largest flooding event in 125 years of data. The hydrodynamic model calculated the water levels around Rio Grande City (MAE ± 0.18 m). These results fed the flooding model, which projected the water over the digital elevation model of the city and produced predictions of flooding conditions on every street (ranging from a few centimeters up to 1.5 m) days before the flooding happened. The results were further customized to attend specific demands from the security forces and municipal civil defense, who evaluated the best alternatives for evacuation strategies and infrastructure safety during the May 2024 extreme flood event. Flood Safety Maps were also generated for all the terminals in the Port of Rio Grande, indicating that the terminals were 0.05 to 2.5 m above the flood level. Overall, this study contributes to a better understanding of the strengths of digital twin models in simulating the impacts of extreme flood events in coastal areas and provides valuable insights into the potential impacts of future climate change in coastal regions, particularly in southern Brazil. This knowledge is crucial for developing targeted strategies to increase regional resilience and sustainability, ensuring that adaptation measures are effectively tailored to anticipated climate impacts. Full article

The article examines the application of artificial intelligence (AI) in the fight against disinformation through a comparative analysis of different European initiatives collected by the SmartVote project. It analyzes their characteristics and contributions to identify common trends in technological development and collaboration models. The methodology combines a systematic documentary analysis of institutional and technical sources—public project records, reports, and official repositories—with a structured questionnaire addressed to the coordinators of the selected initiatives. This mixed approach made it possible to triangulate quantitative and qualitative information on the types of technology employed, areas of impact, stakeholders involved, and levels of funding. The results show a predominance of multimodal AI-based tools aimed at automated content detection and verification. Most of the projects rely on cooperation networks among universities, technology companies, media outlets, and social organizations, structured under the principle of human oversight. The main challenges include algorithmic accuracy, bias prevention, and Europe’s technological dependence. Overall, the initiatives studied are committed to transparency, interdisciplinary collaboration, and the ethical use of AI in defense of informational integrity. Full article

The regulation of sensory processing by centralized neuromodulatory systems can alter behavioral responses to social cues. Neuromodulatory systems such as the serotonergic neurons in the dorsal raphé nucleus (DRN) show heterogenous responses to different types of sensory stimuli or to stimulus qualities such as reward, valence, or salience. Sensory neuromodulation could therefore be related to a broader quality of the behavioral context or to specific types of social cues. We assessed this issue by presenting male mice with either playback of female vocal signals associated with defensive aggression (squeaks) or silence, and the presence or absence of a female. Activity in regions of the DRN that project to the auditory midbrain was assessed through co-labeling with antibodies to the serotonin synthetic enzyme tryptophan hydroxylase (TPH) and the immediate early gene product c-Fos. Female presence or absence had the largest effect, decreasing the co-localization of TPH and c-Fos, while the playback of squeaks had effects that were condition-dependent, increasing co-label only when females were absent. Squeak playback further decreased the correlation in the numbers of co-labeled neurons between two dorsal subdivisions of the DRN, the DRD and DRL. These results are inconsistent with an auditory-exclusive feedback loop. Instead, cues associated with female presence heavily influence raphé activity, with squeaks playing a modifying and context-dependent role. Because the elevation of serotonin in the IC causes males to become more responsive to female squeaks, these findings suggest that a nuanced interaction of positive and negative cues during social interaction may fine-tune male responses to the vocalization of social partners, in part through the serotonergic system. Full article

As China’s urban underground area grows, deep foundation pit projects in complex geological circumstances, particularly near critical infrastructure, must adhere to tight deformation control guidelines. However, limited research has been conducted on the deformation behavior of internal bracing systems in Sichuan’s sandy cobble strata. This research centers on a deep excavation near civil defense facilities in Pujiang County, Chengdu. We investigated the deformation characteristics of retaining piles and internal bracing systems using field monitoring, finite element simulations, and parameter sensitivity analysis, and proposed optimization solutions for the support scheme. Road settlement, pile-head vertical displacement, building settlement, and deep lateral displacement of retaining piles were all monitored in the field at different phases of excavation. MIDAS/GTS was used to generate a 3D finite element model that included bored piles as a contiguous pile wall. The model was verified against monitored data and showed a maximum variation of 3.7%. Parametric studies were conducted to optimize the equivalent stiffness of the contiguous pile wall and the standardized internal bracing system. The findings indicate that the maximum lateral displacement of retaining piles is the primary optimization restriction. Reducing the equivalent stiffness to 0.6t (relative to the baseline thickness t) causes displacement to surpass the warning threshold (35 mm), whereas increasing it to 1.2t or 1.4t limits deformation without incurring significant costs. Case G of the standardized internal bracing system ensures that the maximum pile displacement (21.95 mm) remains below the warning criterion (24.5 mm) while improving constructability. This work elucidates the deformation characteristics of internal bracing systems in sandy cobble strata near sensitive buildings, offering theoretical and practical assistance for comparable projects. Full article

Text-to-image (T2I) generation, a core component of generative artificial intelligence(AI), is increasingly important for creative industries and human–computer interaction. Despite impressive progress in realism and diversity, diffusion models still exhibit critical security blind spots particularly in the Transformer key-value mapping mechanism that underpins cross-modal alignment. Existing backdoor attacks often rely on large-scale data poisoning or extensive fine-tuning, leading to low efficiency and limited stealth. To address these challenges, we propose two efficient backdoor attack methods AttnBackdoor and SemBackdoor grounded in the Transformer’s key-value storage principle. AttnBackdoor injects precise mappings between trigger prompts and target instances by fine-tuning the key-value projection matrices in U-Net cross-attention layers (≈5% of parameters). SemBackdoor establishes semantic-level mappings by editing the text encoder’s MLP projection matrix (≈0.3% of parameters). Both approaches achieve high attack success rates (>90%), with SemBackdoor reaching 98.6% and AttnBackdoor 97.2%. They also reduce parameter updates and training time by 1–2 orders of magnitude compared to prior work while preserving benign generation quality. Our findings reveal dual vulnerabilities at visual and semantic levels and provide a foundation for developing next generation defenses for secure generative AI. Full article