Search Results (1,121)

Background: Rigid alignment of multimodal MRI is a prerequisite for neurological post-processing, yet classical iterative optimisers are accurate only when allowed to converge, at tens of seconds per volume, while learning-based predictors are fast but far less precise. This study asks whether a learned predictor can serve as the initialisation of an iterative optimiser so that both speed and accuracy are obtained. Methods: A supervised dual-path convolutional network with a cross-modal attention block (AttentionReg) was trained on BraTS 2020 under a subject-level split and a synthetic rigid-transformation protocol. Its prediction initialises a brain-masked Mattes Mutual-Information optimiser, forming a guarded hybrid pipeline in which a refinement is accepted only when it improves the metric. Eight methods were compared under one protocol on a common subset of 990 held-out pairs from 99 subjects, using a landmark-based Target Registration Error expressed in physical millimetres. Results: The proposed hybrid attained a median TRE of 0.81 mm over the evaluated pairs (IQR 0.57–1.12; P95 1.67 mm; 0.3% above 3 mm) at 0.79 s per volume, against 0.57 mm at 25.3 s for the fully converged iterative reference. The subject-level paired difference was 0.23 mm (95% CI 0.19–0.28), statistically significant yet equivalent within a 0.50 mm margin, obtained with a 32-fold reduction in computation time. Used alone, the network reached 3.10 mm; iterative refinement improved it by a mean subject-level paired reduction of 2.49 mm (95% CI 2.32–2.67). An ablation showed the attention block does not change median accuracy relative to a matched network without it. Conclusions: A learned initialiser converts an accurate but slow iterative method into an accurate and fast one, providing a proof of concept for real-time rigid multimodal MRI registration. Full article

Large language model (LLM)-powered retrieval-augmented generation (RAG) systems significantly reduce factual errors in question answering, but they present a novel and under-investigated attack surface: factual information in retrieved text can be exposed in ways that do not align with the disclosure norms of the information itself. In legal, medical, and enterprise environments, this leakage is not only a confidentiality violation but a contextual integrity (CI) violation, where privacy is understood as the appropriate flow of information between roles and for specific purposes. Current defenses are mostly input minimizers or post hoc output filters that are blind to the circumstances of the data source (sender, recipient, purpose), creating a structural gap between the document store and the model generation step. We introduce ContextGuard-RAG, a multi-agent privacy enforcement framework that embeds CI theory directly into the retrieval and generation pipeline. It consists of three tightly coupled components: a CI-policy encoder that attaches sender, recipient, subject, information-type, and transmission-principle norms to retrieved contexts from document metadata; a privacy-aware reranker that filters contexts violating inferred CI norms using a fine-tuned cross-encoder classifier; and a generative firewall agent that sanitizes output using reinforcement learning to avoid transitive leakage through document-derived hallucinations or indirect inferences. On PrivacyQA, MedQA, and LegalBench, ContextGuard-RAG reduces CI violations by about 34 percent relative to vanilla RAG, by 13.5 percent on average over AirGapAgent, and by 10 percent on average over the 1-2-3 Check multi-agent reasoning approach, while achieving parity with the best accuracy baseline in ROUGE-L (within 0.6 absolute points) and remaining stable under context-hijacking adversarial probes. Unlike prior approaches, ContextGuard-RAG performs norm-aware retrieval rather than post hoc filtering and benefits four backbone LLMs without retraining the upstream agents. Full article

True-Context (TContext) is a gamified VR platform for fire hazard recognition; empirical evidence shows it outperforms non-contextual VR delivery, but its theoretical positioning relative to established learning frameworks remains unarticulated. This study examined TContext against thirteen learning frameworks to: (1) assess the degree of theoretical alignment between TContext’s design and each framework’s prescriptions; (2) identify significant gaps; and (3) derive design recommendations to strengthen TContext’s theoretical completeness and future development. Using qualitative document analysis of the published TContext corpus and the canonical literature on thirteen learning frameworks, the study examined alignment, divergence, and potential integration. We coded each framework using a directed content analysis approach, then analyzed six cross-framework themes (social/collaborative mediation, individualization/adaptivity, reflective/metacognitive processing, learner agency, transfer/longitudinal validation, and design-inferred versus measured-outcome evidence) to identify recurring patterns. The analysis produced 52 documented strengths and 44 limitations. TContext aligned most strongly with ELT, CLT, CTML, LM-GM, and Constructivism; moderately with FT, Gamification of Learning, LBD, and DT; partially with TPACK and LPS; and least with SDL and Social Constructivism. Four persistent cross-framework gaps emerged: the absence of collaborative learning affordances, the lack of adaptive personalization, the lack of longitudinal retention measurement, and constrained learner agency within pre-scripted scenarios, along with seven targeted design recommendations to address each gap. TContext emerges as a novel multi-framework integrative model whose three core constructs, Contextual Embeddedness, Temporal Stratification, and Distributed Awareness, are not collectively anticipated by any existing framework. Future research should test the recommendations, measure longitudinal retention, and validate the model cross-culturally beyond the UAE. Full article

Reordering operations in a declared finite, straight-line MLIR Transform IR core must preserve request availability, handle phases, effects, and completed outcomes. We present a model-relative certificate framework for this core, restricted to operation handles and matcher-generated carriers. Polynomial interfaces describe request menus, completed coalgebras represent success and failure, and exact branch descriptors compile to ordinary fixed-unit Petri nets. For total-success paths, our reordering theorem constructs the swapped path from residual guards and derives Petri interchange from disjoint compiled supports. With exact atomic source summaries, concrete-to-rich refinement, and an exact target quotient as premises, the two paths have equivalent concrete outcomes. Two worked instances cover unit annotations and guarded handle generation. An independent finite-model audit validates 29 complete-marking transitions and 12 local exchange squares, and rejects seven targeted semantic mutations. Checks of 39 archived MLIR cases in each of two runs confirm the expected payload outputs and diagnostics. A repeated evaluation across 522 files in a fixed corpus available during development constructs 617 V3 kernel records, compared with 292 for V2, and records five native successes among six candidates. A mirror benchmark over 39 conditions confirms the structural count formulas. These results validate the finite model-to-Petri construction and show that the frozen pipeline operates reproducibly on the declared corpus. Full article

Adversarial patch attacks pose a tangible physical-world threat to traffic sign recognition in autonomous driving systems. Current state-of-the-art defenses based on image reconstruction require dual-model deployment and add per-frame inference latency, making them impractical for resource-constrained embedded platforms such as mass-produced ADAS systems and aftermarket dashcams. This paper proposes a two-component defense that separates detection from mitigation. PatchGuard-Freq leverages frequency-domain analysis to detect attacked images with high accuracy, while adversarial fine-tuning enables the detector to recover most of its detection performance under attack without adding any inference module or architectural change. Experimental results show that the approach is effective in the latency regime targeted by this study: on the evaluated stop-sign and nine-class LISA benchmarks, under the evaluated detection placement (a fresh $110\times 110$ patch at a uniformly random position), the detector achieves AUC = 1.0 with zero false positives across all four dataset configurations and all combined frequency variants, while the frequency-only arm of the ablation separates the data only where the patch covers almost the entire input, and adversarial fine-tuning recovers attacked mAP@0.5 on the COCO stop-sign test set from 19.4% to 68.4% on the $110\times 110$-patch configuration while degrading clean mAP by only about 0.7 points. We further show that PatchGuard-Freq is vulnerable to adaptive (defense-aware) attacks and introduce defense-aware training that restores detection of such attacks to 100% without increasing false alarms. This work characterizes the complementary relationship between reconstruction-based and robustness-based paradigms in the accuracy–efficiency design space under the evaluated conditions: the former suits compute-unconstrained scenarios while the latter serves latency-constrained deployments. All quantitative results were obtained on desktop-grade GPUs; embedded-platform latency, memory, and energy were not measured, and the embedded discussion is limited to hardware-independent parameter and FLOP counts. Full article

Spatial interference among the shearer cutting drum, ranging arm, and hydraulic-support face guard is a practical safety risk in fully mechanized coal mining faces, where dust, water mist, weak illumination, and metal multipath reflections limit vision- and LiDAR-based perception. This study develops a 4D millimeter-wave radar point cloud sensing method for face guard segmentation and cutting-interference warning. Radar point clouds were extracted from ROS bag data, converted to frame-level point cloud files, manually labeled, and processed using a joint passthrough-radius filtering strategy. A PointNet++ semantic segmentation network was modified with radar feature augmentation and class-aware neighborhood sampling to identify sparse face guard points from background structures. The segmented face guard points were fitted by an axis-aligned 3D bounding box, and the minimum distance from shearer dangerous regions to the box was used to classify safe, warning, and dangerous states. On a 2160-frame labeled dataset, the final segmentation model reached 99.126% precision, 97.477% recall, and 96.608% IoU at epoch 100. Static pose, dust-condition, and dynamic sequence tests further verified that the proposed sensing pipeline can provide continuous face guard localization and distance-based interference monitoring under representative experimental conditions. Full article

With the rapid development of graph neural networks and physics-informed machine learning for critical infrastructure risk, reliable assessment of submarine telecommunication cables has become both a practical resilience requirement and a demanding cross-domain learning problem. These cables carry more than 99% of international data traffic, yet fault risk modeling faces four explicit challenges: (1) modality misalignment, because marine evidence combines gridded environmental fields with irregular vessel trajectories; (2) relational heterogeneity, because hazards interact through semantically distinct spatial links; (3) physical inconsistency, because unconstrained predictions may violate seabed geomechanics under geographic shift; and (4) decision uncertainty, because safety-critical inspection and routing require uncertainty rather than point estimates alone. The closest approaches leave identifiable gaps. Makrakis and colleagues optimized static cable routes without learned hazard interactions or uncertainty; Taghizadeh and colleagues constrained flood graph predictions without cable-specific heterogeneous entities; and Guo and colleagues fused maritime trajectories without forecasting cable faults or screening routes. No existing approaches combine these missing capabilities under geographically held-out cable basins. To address this gap, the present paper proposes MarineGuard-GNN, a physics-informed multimodal heterogeneous graph neural network. Its Cross-Modal Spatiotemporal Tokenizer maps GEBCO bathymetry, CMEMS ocean fields, and NOAA AIS trajectories into a shared 256-dimensional space; a relation-aware Heterogeneous Graph Transformer represents four semantic node types and four physical relation types; a differentiable Mohr–Coulomb loss regularizes geomechanical consistency; and a Monte Carlo dropout risk head estimates segment-level epistemic uncertainty for inspection and routing. Under 4-fold geographic cross-validation at natural prevalence on 15,110 cable nodes (847 faults), MarineGuard-GNN attains cross-basin AUC-ROC, average-precision, and F1 ranges of $0.72$–$0.76$, $0.18$–$0.24$, and $0.29$–$0.36$, respectively; average precision corresponds to a $3.21$–$4.28\times$ lift over the $0.0561$ no-skill prevalence baseline. Paired basin-stratified bootstrap analysis and Holm-corrected tests confirm improvements over the strongest tabular and graph baselines ($\Delta$AUC-ROC $\ge 0.02$, $\Delta$AP $\ge 0.03$; $p_{\mathrm{adj}}<0.05$). Matched ablation shows that removing the corrected mechanics term reduces AUC-ROC by $0.005$ and average precision by $0.010$ without improving calibration. Across three densely sampled public cable corridors, uncertainty-aware routing reduces mean predicted risk by at least $5\%$ while limiting distance overhead to below $3\%$; the conclusion remains stable for risk thresholds from $0.45$ to $0.60$. These results demonstrate statistically supported cross-basin generalization, establishing the proposed framework as a reproducible decision support method. Full article

The amount of time office workers spend seated poses multiple health risks and has been associated with increased musculoskeletal pain. To evaluate the potential association of discomfort and computer utilization with workstation type, we conducted a retrospective analysis of data collected at a large US company as part of their corporate wellness initiative. Workers had either an electric height-adjustable or a traditional desk for more than a year prior to the data collection phase. Data collected included participants’ discomfort levels and multiple objective measures of computer utilization as captured by the office ergonomic software package, RSI Guard. Data were collected over a year-long period and reported for each participant (n ≈ 10,145). After a year, the percentage of participants reporting no discomfort increased by 2.4%. Active computing time was reviewed for the two categories, with traditional users having 2.95 h per day while the height-adjustable desk users had 3.66 h per day (p < 0.001). This difference in active computing time equates to 43 additional active minutes per day for individuals with access to height-adjustable workstations. Findings suggest that individuals with access to height-adjustable workstations may be associated with increased computer use and negligible changes in discomfort over a one-year period. Full article

Wearable single-lead electrocardiogram (ECG) devices are increasingly used for out-of-clinic rhythm monitoring, but the rigour with which their signal-processing pipelines are validated varies widely across the literature, with most published reports relying on either a single clinical comparison or a single algorithmic benchmark. We propose and apply a multi-tier validation methodology—combining clinical reference comparison, expert-annotated database benchmarking, and a deliberate morphological stress-test—and demonstrate it on the SmartGuard single-lead wearable ECG device and its signal-processing pipeline. The three tiers comprise (i) a within-subject clinical comparison in 31 healthy adult volunteers against a Bionet Cardio 7 limb-lead Lead I reference with cardiologist-confirmed measurements (a two-electrode limb configuration approximating Lead I, not a full diagnostic 12-lead ECG), (ii) algorithmic benchmarking on the Lobachevsky University ECG Database (LUDB, 200 records, expert annotations), and (iii) a robustness stress-test on the QT Database (QTDB, 103 records, hybrid expert annotations). The pipeline integrates discrete wavelet transform-based PQRST delineation with a derivative-based QRS-width refinement, a tangent-method T-wave end estimator, and median-absolute-deviation robust averaging. In the clinical tier, mean within-subject biases were negligible for heart rate, QT, QTc and RR (all p > 0.6) and borderline for QRS duration (+7.3 ms, p = 0.066), with a systematic R-wave amplitude underestimation (−0.15 mV, p < 0.0001). On LUDB, heart-rate and RR-interval agreement were excellent (Pearson r > 0.95) and QRS duration good (r = 0.73, MAE 13.8 ms). On QTDB, performance degraded in line with the increased morphological heterogeneity of the dataset, particularly for QT-related metrics. Despite this minimal mean bias at the population level, individual-record QT and QTc errors showed high dispersion, precluding accurate clinical QT interpretation from single wearable traces in individual patients. Moreover, the resting, healthy-adult design of the clinical tier means its findings cannot be extrapolated to arrhythmic conditions or exercising states. The protocol is described in sufficient detail to serve as a template for evaluating other wearable single-lead ECG devices. Full article

Smart electric vehicle (EV) charging infrastructures are evolving into large-scale cyber-physical Internet of Things (IoT) systems that depend on distributed communication, real-time sensing, and spatially coordinated charging operations. However, their interconnected communication architecture exposes charging stations, EV communication links, and network gateways to coordinated distributed denial-of-service (DDoS) attacks. Existing intrusion detection approaches primarily rely on localized or static traffic analysis and therefore have limited capability to capture spatially distributed and temporally evolving attack behavior. This study proposes the Spatio-Temporal Attack Graph Guard Network (STAG-GuardNet), an unmanned aerial vehicle (UAV)-assisted spatio-temporal attack graph learning framework for DDoS detection and security monitoring in smart EV charging networks. The framework integrates spatiotemporal signal conditioning, telemetry-adaptive graph aggregation, temporal dependency learning, attack-memory encoding, and adaptive risk-aware attention to model coordinated cyber-physical attack behavior. UAV-assisted telemetry provides complementary spatial and wireless information on communication instability, signal variation, neighboring congestion, and distributed attack-related behavior. A Hybrid Hawk–Manta Adaptive Optimizer (HHMAO) is employed to improve hyperparameter selection and convergence stability under imbalanced, heterogeneous, and nonstationary traffic conditions. The framework is evaluated using a smart-city EV charging cybersecurity dataset and three benchmark IoT intrusion detection datasets, namely TON_IoT, Edge-IIoTset, and X-IIoTID. Experimental results show that STAG-GuardNet achieves 97.7% accuracy, a 97.7% weighted F1-score, and a 98.4% area under the receiver operating characteristic curve (AUC) on the primary dataset. The framework also maintains stable performance under noisy telemetry, missing observations, heterogeneous traffic distributions, and charging-node outages. These findings demonstrate the potential of STAG-GuardNet for resilient and spatially informed security monitoring in UAV-assisted IoT-enabled EV charging infrastructures. Full article

Correct packet delivery does not guarantee correct semantic interpretation when endpoint meanings for the same learned codeword diverge. CoSMA DAI is proposed for mismatch detection, protected confirmation, task preservation and semantic repair. Channel-conditioned global evidence, semantic class local evidence, temporal dynamics, channel context and protected probe evidence are fused by a causal machine-learned mismatch belief. A transmitter-derived task belief preserves the downstream decision while repair is pending and BDIx agents select guarded intentions for probing, fallback and resynchronisation. Evaluation uses 30 held-out drift seeds, 300 matched null streams and 300 degrading channel controls. Six referenced sequential monitors receive the same conditioned payload score. CoSMA DAI obtains 100.00 percent balanced accuracy, precision, recall, F1 score and Matthews correlation coefficient with zero observed matched null false alarms. Its aggregate delay is 5.62 slots, compared with 11.58 slots for the other zero false alarm method. The task-preservation belief maintains 94.73 percent task accuracy through every divergence scenario, above the quantised accuracy ceiling of 0.919 of the semantic path, because it is derived from the unquantised transmitter latent. A task-label-only control confirms that this accuracy is secured by the preservation belief alone, independently of the detector, so task preservation and mismatch detection are decoupled by design and detectors are compared on residual functional semantic outage, outage duration and semantic reconstruction fidelity, which measure the restoration of the semantic representation itself. Without repair, the residual semantic outage is 73.69 percent at 15.97 dB reconstruction fidelity, whereas CoSMA DAI reduces it to 0.73 percent over 6.62 slots at 21.74 dB. Under five declared parity tiers, in which multivariate and supervised baselines receive the identical features, training seeds, protected probe and candidate budget, the protected confirmation stage reduces false repair for every detector to which it is attached. Zero-shot evaluation over 7 unseen mismatch families and 5 unseen link models retains full detection with zero observed false repair in 6 of the 7 families and on every link and identifies receiver-side decoder drift as a condition the present observation model cannot detect. The learned belief is validated at slot level with an area under the receiver operating characteristic curve of 0.99997 and a class overlap of 0.00039, leave-one-mechanism-out and cross-channel retraining are reported, behaviour is characterised down to the practical detection boundary and scaling to 64-dimensional representations with 2048-entry codebooks is demonstrated. The task-belief mechanism is shown to be economical only for small closed-set output spaces and the channel-conditioning tables are shown to reduce to 6 cells without loss. Every comparator is additionally retuned on the same development budget, paired bootstrap intervals and signed-rank tests are reported over the shared streams, auxiliary traffic and radio energy are normalised per correct decision, authentication of the task belief is specified and charged and transfer to MNIST, Fashion-MNIST, CIFAR-10 and CIFAR-100 is demonstrated without retraining, including on a convolutional VQ-VAE representation with a jointly learned 512-entry codebook, where foreground segmentation and localisation are restored to within the quantisation limit while a class decision cannot serve either task. The control traffic share is 23.59 percent, which is 12.62 percent lower than the monitor value. The additional semantic side information increases radio energy to 0.393 mJ per stream and reduces control-adjusted resource efficiency to 6.203 source-equivalent bits per channel use. The results therefore establish reliable detection and semantic repair within the principal comparison, with comparator-specific delay advantages and without claiming task-accuracy, semantic-rate or energy superiority. Full article

Adaptive cruise control (ACC) tracks a time-gap reference that trades efficiency against spacing safety; in practice this gap is fixed offline, although the safest choice depends on the closing speed and on the reliability of the perception that supplies the lead-vehicle state. We present GL-TGS-v2, a guarded, interpretable supervisor that selects the time-gap reference of an adaptive model predictive controller (MPC) online from tracked lead-vehicle kinematics. The supervisor is a decision-tree distillation of a calibrated adaptive expert, wrapped by a perception-aware safety shield; its only authority is the time-gap reference, so it retrofits onto an existing controller without re-opening the inner loop. We evaluate it in closed loop on a frozen stack that couples a YOLO11 detector, a joint probabilistic data association (JPDA) tracker, and the adaptive MPC, with documented runtime and per-run provenance gates. In a hard lead-braking scenario (ten runs per method), GL-TGS-v2 matches the expert at a matched mean time gap and improves minimum relative distance over fuzzy and fixed-gap baselines by 3.3–5.2 m, with 95% confidence intervals excluding zero, lower target loss, and no collisions. Benign testing shows no regression, while shield ablation characterizes the guard as an auditable protective override. Full article

Artificial intelligence (AI) recommender systems operationalize personalized marketing by transforming behavioral data into individualized choice architectures, yet greater model complexity or personalization granularity need not improve every service objective. This study compares population-level item-neighborhood recommendation, behaviorally segmented neighborhood recommendation, and neural collaborative filtering (NCF) across 776,741 deduplicated 1–5 ratings from 646,576 anonymized users and 541 Coursera courses. The primary warm-start evaluation is restricted to 10,296 users (1.59% of the user population) with sufficient interaction history; the remaining sparse-history users contribute to fitting but not to the within-user confirmatory estimand. Configurations are selected only on three validation seeds and then frozen before guarded final evaluation across ten deterministic user-aware seeds. No rank-capable model dominates across objectives: KNN provides the strongest RMSE and personalized-neighborhood coverage profile, ClusteredKNN achieves the strongest observed Top-K retrieval with lower neighborhood support, and a metadata-rich NCF variant achieves the lowest MAE at substantially greater fitting cost. The rating-only UserMean baseline further shows that low numerical prediction error need not imply useful item ranking. Overall, the findings support a contingency view of AI personalization: greater segmentation granularity or model complexity does not inherently create greater value; model choice should instead reflect the service objective, available behavioral evidence, coverage tolerance, operating cadence, and governance requirements. Full article

Consent banners give digital consumers a right of refusal that websites do not reliably honor. A dual-persona audit of 100 highly trafficked domains across five consumer-facing sectors (news, e-commerce, health, finance, and government) found third-party cookies set before any interaction on 33.0% of domains; no programmatically reachable rejection path on 69%; and, where rejection was possible, third-party cookies continued to be stored on 51.6%, and identifier-pattern cookies were retained on 74.2%. Guided by these findings, we present CyberGuard, a Manifest V3 browser extension that enforces privacy at the client through a minimal network blocklist; fingerprint fuzzing across canvas, WebGL, and audio; behavioral masking; and a human-in-the-loop cookie vault. In a replication matched by browser mode on the same corpus, run on two separate days with all cookies accepted in both conditions, CyberGuard reduced third-party cookies by 56.5% on average (11.3 to 4.9 per domain; Wilcoxon W = 66.5, p = 1.7 × 10⁻¹¹). The initial campaign, in which the two conditions ran under different browser modes, had measured 70.8% (15.37 to 4.49; W = 20.0, p = 1.16 × 10⁻⁵); part of that figure reflects the mode difference rather than the extension. The reduction is structural: its pattern is consistent with a single blocked advertising entry point suppressing a downstream header-bidding stratum far beyond the two targeted networks, while platform embeds largely persisted. Client-side enforcement thus lowers tracking exposure even when consent is granted; the datasets, the consent selectors, and the crawler harness are released. Full article

Steganography using deep learning can preserve pixel-level image quality while still changing object, attribute, or relational information in captions generated by vision-language models (VLMs). This caption drift creates a detection channel that is not measured by global image-embedding similarity alone. This paper presents StegoGuard, a framework that embeds secret payloads while enforcing caption-level semantic consistency. Concept drift is defined as a measurable divergence in object-level, attribute-level, or relational semantics between cover and stego captions and is quantified using CLIP text-embedding cosine distance. The main technical contribution is a cross-modal semantic drift regularization term based on BLIP-2 captions generated for cover and stego images. The framework combines this objective with CLIP-based saliency-guided region selection and a lightweight Vision Transformer encoder-decoder. Saliency-map quality is evaluated against ground-truth segmentation masks, and a deterministic bit-to-patch mapping protocol is provided for reproducibility. Experiments use COCO2017, DIV2K, and BOSSBase. Within the controlled six-baseline protocol, StegoGuard achieved a PSNR of 38.9 dB, an SSIM of 0.976 at 256 bits, detector AUC values of 0.521–0.562, and caption similarity of 0.962 as measured by CLIP text cosine similarity (model-relative, not human-verified). The ablation results show that the drift term reduces the measured concept-shift rates while preserving the reported bit-recovery and image-quality levels. Full article