Search Results (171)

Ocular ductions deform the optic disc and peripapillary blood vessels, and deformations can be interpreted as mechanical strain. We used confocal scanning laser ophthalmoscopy (cSLO) to map strain in disc and peripapillary retinal vessels associated with the cardiac pulse and determine if such strain is influenced by gaze direction. Sets of 13 infrared cSLO images were obtained sequentially for each eye using a Heidelberg Spectralis scanner in cinematic mode over a 3 sec interval in adults. Imaging was repeated in central, and horizontally (30° adduction/abduction) and vertically eccentric gazes (10° supraduction/infraduction). Retinal vessels, optic disc, and fovea were segmented using custom-trained, deep learning-based models. Frame to frame vascular displacements were automatically determined using optical flow analysis, allowing computation of equivalent strain. A total of 25 eyes of 13 subjects of mean age 39 ± 18 (standard deviation, range: 25 to 81) years were included. Average equivalent strain over 3 sec ranging from 0.27% to 0.36% exceeded the 0.16% noise threshold across all gazes and regions, indicating measurable pulse-induced deformation. After adjustment for age and axial length, pulsatile maximum and minimum strain were influenced slightly by gaze direction, maximally for supraduction, whereas mean strain did not vary significantly with gaze. The cardiac pulse induces measurable deformation of retinal vessels that can be quantified as equivalent strain in the image plane using optical flow-derived displacement fields. However, the interaction of pulse strain with gaze direction is unlikely to be a significant confound for investigations of strains associated with eye movements. Full article

Operational Technology (OT) and Industrial Control Systems (ICSs) along the NATO-EU eastern flank face escalating hybrid threats, yet existing cyber-resilience metrics remain IT-centric, lacking OT-specific constraints and geopolitical exposure dimensions. This paper presents a Design Science Research contribution: the development and simulation-based feasibility demonstration of two interconnected artefacts. The first is the AI-enabled Cyber-Resilience Index (ACRI)—a composite 0–100 metric operationalized through 16 indicators across four domains (detection performance, operational continuity, governance maturity, supply-chain risk), aggregated as a three-term convex combination of capability domains with a linear subtractive supply-chain exposure penalty, weighted via AHP-based illustrative sector-reference profiles. The second is the Unified Compliance Framework (UCF), a structured R → C → E → SLO mapping linking 47 atomic regulatory requirements (NIS2, DORA, CER, AI Act, CRA) to standards (IEC 62443, ISO/IEC 27001) and auditable evidence artifacts, with a Continuous Assurance Loop operationalizing continuous control monitoring. Feasibility is demonstrated through digital twin simulation under three OT-representative threat scenarios (energy SCADA APT, railway supply-chain compromise, manufacturing ransomware). Results in simulated environments show ACRI improvement from Moderate-Risk baselines (45–61) to Adequate-Resilience thresholds (65–73); the proposed federated autoencoder–LSTM detector attains a composite $D_{perf}$ of 0.883 versus 0.510 for a static ±3$\sigma$ threshold baseline (a 73% relative improvement at the domain level). Sensitivity analysis confirms classification robustness ($\pm 7.3\%$ weight perturbation; coefficient of variation below 9.1% across 10,000 Monte Carlo iterations). Critical limitations are explicit: simulation-only evidence ($n=12$ scenario instances), illustrative (non-empirical) AHP weights, no operational field validation, and limited inferential statistical power. instances), illustrative (non-empirical) AHP weights, no operational field validation, and limited inferential statistical power. The contribution is positioned as a proof-of-concept design artifact establishing methodological foundations for OT-centric resilience assessment and compliance-to-engineering traceability, not as a field-validated operational system. Full article

Large language model (LLM) serving has emerged as a major source of energy consumption in modern AI infrastructure. In current deployments, graphics processing units (GPUs) are typically operated at default high-frequency settings to maximize performance. However, under practical service-level objectives (SLOs), peak performance is often unnecessary, especially during the memory-bound decode stage, resulting in substantial power redundancy and avoidable energy waste. Existing studies that apply GPU dynamic voltage and frequency scaling (DVFS) to improve the energy efficiency of LLM serving have shown promising results. However, they generally rely on coarse-grained control, accurate output length prediction, or request-level resource management, which limits their effectiveness under highly dynamic workloads and strict SLO constraints. We present EcoInfer, a fine-grained DVFS framework for energy-efficient LLM serving. EcoInfer performs iteration-level, workload-aware GPU frequency control that adapts to the current inference phase and system state while preserving latency guarantees. It comprises three tightly integrated modules: a machine-learning-based frequency–latency predictor that estimates iteration latency across candidate GPU frequencies using lightweight iteration-level features; an SLO-aware frequency controller that selects the minimum feasible frequency within a sweet-spot-guided candidate range; and a low-overhead runtime optimization layer that combines adaptive decision caching with asynchronous execution to reduce and hide the overhead of online control. Implemented on top of vLLM, EcoInfer achieves up to 25.4% energy savings and 21.5% average energy savings and improves energy efficiency by 1.28× on average in terms of Tokens/J while maintaining a nearly unchanged SLO attainment rate compared with the default vLLM baseline. Full article

Elastic cloud systems are increasingly employing machine learning (ML) to automate resource scaling in response to variable workloads and stringent service-level objectives. However, current ML-based autoscalers are fragmented across different platforms, objectives, and evaluation frameworks. This survey examines 60 primary studies conducted between 2015 and 2025, categorising them according to a five-dimensional taxonomy that includes goal, decision logic, scaling mode, control scope, and deployment. This study classifies supervised, unsupervised, and reinforcement learning approaches and analyzes their integration into practical frameworks, including Kubernetes-based controllers and cloud provider services. This paper summarizes the application of machine learning to workload prediction, proactive and hybrid horizontal–vertical scaling, and adaptive policy optimization. Additionally, it synthesises common evaluation practices, encompassing workloads, metrics, and benchmarks. The analysis identifies ongoing challenges: actuation delays and telemetry lag, the intricacies of hybrid scaling, coordination across multi-service and edge-cloud deployments, and the constrained joint consideration of cost, SLO, and energy objectives. The identified gaps necessitate additional research on unified machine learning-driven orchestration, multi-agent and federated control, standardised benchmarks, and sustainability-aware autoscaling. Full article

Background: Group A Streptococcus (GAS) induces a wide spectrum of human diseases, ranging from superficial infections to life-threatening invasive conditions and post-infectious sequelae such as rheumatic heart disease, posing a heavy global health burden. Critically, there is still no licensed commercial vaccine against GAS, making the development of novel, effective vaccines against this pathogen an urgent and crucial unmet medical need. Methods: We developed a dual-antigen nanoconjugate vaccine against GAS. The Group A Carbohydrate polyrhamnose backbone (GAC^PR) and truncated SLO were site-specifically conjugated via Protein Glycan Coupling Technology (PGCT) in engineered E. coli, and then linked to ferritin nanoparticles using the SnoopTag/SnoopCatcher system. Safety, immunogenicity, and protective efficacy were evaluated in murine models. Results: The nanovaccine was successfully synthesized with high purity. It elicited robust GAC- and SLO-specific IgG/IgG1 responses, conferred 90% survival against lethal GAS challenge (vs. 0–50% in controls), reduced bacterial loads in organs, and lowered inflammatory cytokines. Passive immunization with vaccine-induced serum also achieved 90% survival. No abnormal biochemical indicators, inflammatory responses, or organ pathology were observed. Conclusions: This study successfully developed a bivalent nanoparticle vaccine against GAS. This novel nanovaccine exhibits excellent safety, strong immunogenicity, and effective protection against GAS, providing a promising vaccine candidate. Full article

Propose an enhanced LoRA (Low-Rank Adaptation) MoE (mixed expert) architecture, SLoRA (Enhanced LoRA MoE Architecture), aimed at addressing the key problem of efficient parameter fine-tuning in multitasking scenarios. Given the high cost of traditional full fine-tuning as the parameter size of visual language models increases, and the limitations of LoRA as a popular PEFT (parameter-efficient fine-tuning) method in multitasking, such as inadequate adaptability and difficulty in capturing complex task patterns, as well as the catastrophic forgetting and knowledge fragmentation challenges faced by existing research on integrating mixed expert (MoE) mechanisms into LoRA, SLoRA utilizes orthogonal constraint optimization to reduce disturbance to existing knowledge through constraint solution space initialization, alleviating catastrophic forgetting (old task accuracy retention rate reaches 92.4%, 16.1% higher than LoRA), and an optimized MoE structure that includes general experts (retaining pre-trained knowledge) and task-specific experts (dynamic routing adaptation tasks) to enhance multitask adaptability. Experimental results show that in commonsense reasoning tasks, SLoRA’s accuracy is 9.0% higher than LoRA and 3.7% higher than AdaLoRA on the WSC dataset, and its F1 score is 7.7% higher than LoRA and 2.9% higher than AdaLoRA on the CommonsenseQA dataset; in multimodal tasks, its average score is up to 15.3% higher than LoRA, demonstrating significant advantages over existing methods. Full article

Large language models in IoT–edge–cloud settings face bursty, heterogeneous requests that make pipeline-parallel inference prone to micro-batch imbalance and communication stalls, causing GPU idle time and SLO violations. We propose a runtime-adaptive scheduler that jointly tunes token budgets and micro-batch counts to balance prefill/decode workloads and minimize pipeline bubbles under changing compute and network conditions. On a four-node pipeline-parallel cluster across Llama-2-13b and Qwen2.5-14b at 100/1000 Mbps, our method outperforms vLLM and SGLang, reducing GPU idle time by up to 55% and improving throughput by up to 1.61 × while improving TTFT/ITL SLO satisfaction. These results show that dynamic scheduling is essential for scalable, latency-stable LLM inference in IoT–edge–cloud environments. Full article

Background: Smith–Lemli–Opitz syndrome (SLOS) is an inborn error of cholesterol biosynthesis, caused by biallelic mutations in the DHCR7 gene. Genotype–phenotype correlations regarding DHCR7 variants could explain the variation in severity, ranging from in utero demise or severe SLOS to a mild phenotype. Clinical recognition can be challenging. This study aimed to determine the frequency of SLOS carriers in the Central European population, as well as the mutational spectrum of DHCR7 in these carriers. Methods: A retrospective analysis of DHCR7 variants was conducted using next-generation sequencing data from 55,289 individuals in Czech and Hungarian genetic laboratories. Results: The SLOS carrier frequency and the mutational spectrum of the DHCR7 gene in its carriers were established in the Czech and Hungarian sub-cohorts. In the combined dataset, we identified causative DHCR7 variants on 1567 alleles among 55,289 tested individuals, contributing to an SLOS carrier frequency of 2.83%. Of the 31 DHCR7 variants detected, the c.452G>A variant was the most prevalent, accounting for 1.8% of all detected alleles in our cohorts. In contrast, the c.964-1G>C variant was more frequent in non-Finnish Europeans, as indicated by the gnomAD 4.1.0 database. The DHCR7 mutational spectra of patients and carriers were comparable in terms of the most common variants. Conclusions: The high SLOS carrier frequency (2.83%) underscores the importance of SLOS carrier screening in Central European populations. The prevalent DHCR7 null mutations and their potential combinations may explain the lower-than-expected prevalence of SLOS, whilst Central and Eastern European populations remain likely underrepresented in the current gnomAD database. Full article

The rapid proliferation of connected vehicles equipped with both Vehicle-to-Vehicle (V2V) sidelink and cellular interfaces creates new opportunities for real-time vehicular applications, yet achieving ultra-reliable communication without prohibitive cellular costs remains challenging. This paper addresses reliable inter-vehicle video streaming for safety-critical applications such as See-Through for Passing and Obstructed View Assist, which require stringent Service Level Objectives (SLOs) of 50 ms latency with 99% reliability. Through measurements in Seoul urban environments, we characterize the complementary nature of V2V and Vehicle-to-Network-to-Vehicle (V2N2V) paths: V2V provides ultra-low latency (mean 2.99 ms) but imperfect reliability (95.77%), while V2N2V achieves perfect reliability but exhibits high latency variability ($P_{99}$: 120.33 ms in centralized routing) that violates target SLOs. We propose a hybrid framework that exploits V2V as the primary path while selectively retransmitting only lost packets via V2N2V. The key innovation is a dual loss detection mechanism combining gap-based and timeout-based triggers leveraging Real-Time Protocol (RTP) headers for both immediate response and comprehensive coverage. Trace-driven simulation demonstrates that the proposed framework achieves a 99.96% packet reception rate and 99.71% frame playback ratio, approaching lossless transmission while maintaining cellular utilization at only 5.54%, which is merely 0.84 percentage points above the V2V loss rate. This represents a $7\times$ cost reduction versus PLR Switching (4.2 GB vs. 28 GB monthly) while reducing video stalls by $10\times$. These results demonstrate that packet-level selective redundancy enables cost-effective ultra-reliable V2X communication at scale. Full article

Data-driven quality control (QC) systems for the hot forming of steel parts increasingly rely on deep learning models deployed at the network edge, making multivariate sensor time series a critical asset for both local decisions and management information system (MIS) reporting. However, these models are vulnerable to adversarial perturbations and realistic signal disturbances, which can induce misclassification and distort key performance indicators (KPIs) such as first-pass yield (FPY), scrap-related losses, and latency service-level objectives (SLOs). To address this risk, this study introduces a Digital-Twin-Conditioned Diffusion Purification (DTCDP) framework that constrains latent diffusion-based denoising using process states from a lightweight digital twin of the hot-forming line. At each reverse-denoising step, the twin provides physics residuals that are converted into a scalar penalty, and the diffusion latent is updated with a guidance term. This directly bends the sampling trajectory toward reconstructions that adhere to process constraints while removing adversarial perturbations. DTCDP operates as an edge-side preprocessing module that purifies sensor sequences before they are consumed by existing long short-term memory (LSTM)-based QC models, while exposing purification metadata and physics-guidance diagnostics to the plant MIS. In a four-week production dataset comprising more than 40,000 bars, with white-box ℓ∞ attacks crafted on multivariate sensor time series using Fast Gradient Sign Method and Projected Gradient Descent at perturbation budgets of 1–3% of the physical range, combined with additional realistic disturbances, DTCDP improves the robust classification performance of an LSTM-based QC model from 61.0% to 81.5% robust accuracy, while keeping clean accuracy (≈93%) and FPY on clean data (≈97%) essentially unchanged. These results indicate that physics-aware, digital-twin-guided diffusion purification can enhance the adversarial robustness of edge QC in hot forming without compromising operational KPIs. Full article

Background/Objectives: Smith–Lemli–Opitz syndrome (SLOS) is a genetic metabolic disorder characterized by impaired cholesterol synthesis and a wide range of developmental anomalies. This article presents a case of a girl with SLOS, diagnosed with two pathogenic variants of the DHCR7 gene. The objective is to evaluate the impact of early, multidisciplinary therapeutic interventions on the patient’s development. Methods: Following diagnosis, a comprehensive metabolic therapy was initiated, including cholesterol and cholic acid supplementation. An interdisciplinary therapeutic approach was employed, involving physical therapy, speech therapy, and sensory integration, aimed at addressing various developmental challenges faced by the patient. Results: The therapy led to gradual improvements in the patient’s psychomotor development, although the cholesterol levels were only partially improved and the accumulation of sterol precursors (7-DHC and 8-DHC) persisted. The coordinated care model facilitated better outcomes compared to less integrated efforts. Conclusions: The results highlight the importance of early diagnosis and integrated care in optimizing developmental outcomes for children with SLOS. A multidisciplinary approach is essential for addressing the complexities of the syndrome and promoting overall development. Full article

Student learning outcome (SLO) tracing aims to monitor students’ learning progress by predicting their likelihood of passing or failing courses using Deep Knowledge Tracing (DKT). However, conventional DKT models often lack interpretability, limiting their adoption in educational settings that require transparent decision-making. To address this challenge, this quantitative study proposes an interpretable ensemble framework that integrates Item Response Theory (IRT) with DKT. Specifically, multiple IRT-based DKT models are developed to capture student ability and item characteristics, and these models are combined using a bagging strategy to enhance predictive performance and robustness. The framework is evaluated on an SLO tracing dataset from Prairie View A&M University (PVAMU), a historically Black college and university (HBCU). Result analysis includes comparisons of evaluation metrics such as Area Under the Curve (AUC), accuracy (ACC), and precision across individual and ensemble models, as well as visualizations of student ability, item difficulty, and predicted probabilities to assess interpretability. Experimental results demonstrate that the ensemble approach consistently outperforms single models while providing clear, interpretable insights into student learning dynamics. These findings suggest that integrating ensemble methods with IRT can simultaneously improve prediction accuracy and transparency in SLO tracing. Full article

Inference now dominates the lifecycle footprint of large language models, yet published estimates often use inconsistent boundaries and optimize carbon while ignoring water. We present a provider-agnostic framework that unifies scope-transparent measurement with time-resolved, SLO-aware orchestration and jointly optimizes carbon and consumptive water. Measurement reports daily medians at a comprehensive serving boundary that includes accelerators, host CPU/DRAM, provisioned idle, and PUE uplift, and provides accelerator-only whiskers for reconciliation. Optimization uses a mixed-integer linear program solved over five-minute windows; it selects region, batch size, and phase-aware hardware for prefill and decode while enforcing $p95$ TTFT and TPOT as well as capacity constraints. Applied to four representative models, a single SLO-aware policy reduces comprehensive-boundary medians by 57 to 59 percent for energy, 59 to 60 percent for water, and 78 to 80 percent for location-based CO₂, with SLOs met in every window. For a day with 500 million queries on GPT-4o, totals fall from 0.344 to 0.145 GWh, 1.196 to 0.490 ML, and 121 to 25 t CO₂ (location-based). The framework offers a deployable template for carbon- and water-aware LLM serving with auditable and scope-transparent reporting. Full article

Oil and its derivatives affect marine ecosystems due to pollution. Analytical methods for detecting oils and greases in saline water can identify oil-derived pollutants in seas and oceans, supporting the preservation and recovery of water quality. This study describes a methodology based on Raman spectroscopy to quantify oil in saline water. Specific seriate volumes of synthetic lubricating oil (SLO) and diesel fuel oil (DFO) were added to a beaker containing 1000 mL of saline water. A magnetic stirrer was used to create vortex, where the added oil dispersed uniformly over the surface and created a thin film. Raman spectra of the surface’s film were obtained by a spectrometer (830 nm, 350 mW) at a fixed position with reference to the beaker border, in triplicate. Two spectral models were developed; one based on the intensity of the peak at ~1400–1500 cm⁻¹ and another based on partial least squares regression (PLSR). Both spectral models enabled the quantification of SLO and DFO at concentrations ranging from 25.6 to 307 mg/L, and from 16.8 to 205 mg/L, respectively, with correlation coefficients as high as r = 0.99. The results highlight the potential of using Raman spectroscopy for analyzing oil in environmental water samples. Full article

Objectives: The aim of this study was to determine which color imaging facilitated easier detection of the persistent avascular retina (PAR) in ultra-widefield (UWF) fundus imaging in children undergoing retinopathy of prematurity (ROP). Methods: A total of 20 eyes of 10 children aged between 6 and 9 who underwent diagnostic and therapeutic procedures for ROP were included. Fundus images were obtained using Optos confocal scanning laser ophthalmoscopy (cSLO; Optos PLC, Daytona, Dunfermline, UK). The images were divided and recorded into three groups as original imaging (composite), red reflectance imaging, and green reflectance imaging. These images were prepared as a slideshow for 10 ophthalmology specialists and they were surveyed to determine in which color imaging the peripheral avascular area was more easily detected. The results were evaluated. Results: The rate of detecting the PAR in green reflectance imaging by the participants included in the study was found to be statistically higher compared to other colors of imaging (composite 0.63 ± 0.09 (0.5–0.8), red 0.12 ± 0.05 (0.05–0.2), and green 0.94 ± 0.06 (0.85–1), p < 0.0001). All respondents reported that the boundaries of the peripheral avascular area were more clearly defined in the UWF (Optos PLC, Daytona, Dunfermline, UK) green reflectance imaging. Conclusions: Each color imaging used in UWF fundus imaging helps to visualize different layers of the retina. Our study showed that retinal vascular endings appear more distinct due to the lower penetration of the green laser into the choroidal vessels. Based on these findings, we believe that UWF fundus green reflectance imaging is more useful for detecting and monitoring PAR. Full article