Search Results (346)

Snapshot compressive multi-view spectral imaging (SC-MVSI) multiplexes view-spectral information into a single coded measurement, enabling compact acquisition with a two-dimensional detector. Because each reconstructed channel corresponds to both a selected spectral response and a view direction, direct cross-channel modeling at identical pixel coordinates can introduce structural mismatch caused by view-dependent displacement. This paper proposes a reference-guided view-aligned nonlocal low-rank tensor reconstruction method for SC-MVSI. The reconstruction is formulated as a coded inverse problem and solved using the alternating direction method of multipliers (ADMM) in a variable-splitting framework. In the prior update, a reference tensor guides block-level patch alignment before nonlocal tensor grouping, and the resulting fourth-order tensor groups are regularized by canonical polyadic (CP) low-rank approximation. Experiments on eight synthesized multispectral light-field scenes show that the proposed method achieves the highest average PSNR of 33.61 dB and the lowest average CAE of 5.69 degrees among the compared baselines, while obtaining the second-highest average SSIM of 0.8823. Real-system experiments further provide a qualitative demonstration of applying the proposed reconstruction framework to captured coded measurements. Full article

Radio frequency interference (RFI) caused by spectrum allocation issues can adversely affect the normal operation of synthetic aperture radar (SAR). Compared to traditional RFI, pulsed RFI (PRFI) typically has higher power and wider bandwidth, making the resulting artifacts more likely to obscure the targets of interest. The notch filtering methods are robust approaches to mitigating PRFI. However, the proportion of PRFI in SAR data increases as the electromagnetic environment gradually deteriorates, and these methods inevitably lose too much useful signal. Moreover, traditional semi-parametric methods based on nuclear norms (NNs) over-penalize singular values, leading to the inaccurate estimation of low-rank components. To overcome the aforementioned problems, this article proposes a semi-parametric method. Our method is based on the Hankel structure and truncated nuclear norm (TNN) constrained low-rank estimation model to mitigate strong PRFI while preserving valuable signals. First, the SAR echoes containing PRFI are detected using a relative energy ratio algorithm. Then, the low-rank properties of PRFI are linearly expanded and enhanced through the Hankel structure. Finally, TNN regularization is employed to obtain more accurate low-rank approximations than traditional NN, thereby separating PRFI. Experiments based on simulated and measured data from ESA Sentinel-1A validated the usefulness and preeminence of our method. Full article

Training-free LLM compression avoids fine-tuning by approximating weight matrices from a small unlabelled calibration set; however, existing methods assign each layer an independent target rank with no global budget coordination, leaving parameter distribution across layers systematically suboptimal. We propose CLEAR (Convex-optimal Layer Energy Allocation and Representation), a training-free compression framework that combines an activation-whitened structured-dictionary representation with a provably optimal, globally coordinated budget allocation across layers. Ranks are distributed across all layers simultaneously via a convex water-filling optimization, whose KKT solution is provably optimal and achieves <${10}^{-3}$ pp precision on the dictionary-path parameter budget (the total realized retention ratio, including EoRA and outlier bypass parameters, deviates from the target by at most 0.02 pp in practice). Each layer is then compressed using activation-whitened structured sparse dictionaries ($W\approx DC$, k-sparse columns), supplemented by an EoRA low-rank residual correction and an outlier input-channel bypass. Evaluated on five LLMs spanning 600 M to 8 B parameters across four model families, CLEAR outperforms the current state of the art on three of four evaluated models at compression ratio (CR) = 0.2. On LLaMA-3.2-1B, average accuracy across eight zero-shot benchmarks reaches 50.4% versus 42.7% for CoSpaDi and 37.6% for SVD-LLM, with perplexity reduced from 63.7 to 22.4; on LLaMA-3-8B, 65.5% versus 61.8% for CoSpaDi. Component ablations confirm +5.22 pp from sparse dictionaries and +1.12 pp from global rank allocation. KFAC Fisher covariance proves counterproductive in this setting, degrading accuracy by 8.1 pp due to numerical overflow in SiLU gating layers and consequent budget misallocation. CLEAR completes in 8–30 min on a single GPU, requires no gradient computation at any stage (forward-pass only, including the cascade activation refresh in Phase 3), and reduces peak deployment memory by approximately 20% at BFloat16 precision. Full article

Hyperspectral image (HSI) denoising remains a critical challenge due to noise corruption during acquisition. While nonlocal low-rank (LR) tensor methods leverage spatial–spectral correlations, they usually fail under heavy or complex noise, as directly estimating LR tensors from noisy observations usually leads to residual noise accumulation. To address this limitation, we propose a novel nonlocal low-rank residual (NLRR) approach, which reformulates LR tensor recovery as a progressive residual minimization problem. Unlike conventional methods that exclusively approximate LR tensors directly from degraded observations, the proposed NLRR approach iteratively refines the latent LR tensor structure by minimizing the rank residual, thereby decoupling noise suppression from tensor approximation. This residual-driven framework uniquely integrates two complementary priors: (1) a nonlocal LR residual prior that exploits spatial self-similarity, and (2) a global spectral LR prior that suppresses spectral redundancy. To generalize the proposed NLRR approach to real-world scenarios with mixed noise, we develop the NLRR-robust principal component analysis (NLRR-RPCA) framework, which incorporates the LR residual along with global spectral LR and sparse tensor priors for mixed noise removal. Additionally, to ensure both numerical stability and computational tractability, we develop an adaptive rank-adjusted alternating minimization algorithm, which dynamically adjusts the ranks of the estimated tensors to better handle different noise scenarios. Extensive experiments on both simulated and real HSI datasets demonstrate that our proposed NLRR approach outperforms numerous popular or state-of-the-art methods in both quantitative evaluation and visual perception. Full article

Future sixth-generation (6G)-oriented networks require programmable control that can adapt routing to latency and congestion without unsafe online exploration. This study evaluates offline multi-agent deep deterministic policy gradient (MADDPG) with behavior-adjusted training rewards for latency-aware path control in software-defined networking (SDN). Each traffic pair is modeled as an agent selecting one of three retained candidate paths, while centralized critics learn coordinated decisions from topology-specific Ryu–Mininet transition datasets. Nine policies are compared using ten paired seeds on fat-tree, mesh-grid, and WAN-corridors topologies under a deployed utilization–latency weighting of $0.60/0.40$, together with flow-completion, latency, congestion, architectural-comparison, sensitivity, robustness, statistical, and controller-overhead analyses. The utilization-aware path heuristic achieves the strongest overall reward ranking. MADDPG is the strongest learned policy on fat-tree, is not significantly outperformed by any evaluated policy on mesh-grid, and remains statistically tied with completion-matched policies on WAN-corridors. Behavior adjustment is topology-dependent rather than uniformly beneficial. The exported policy requires approximately $52\mathsf{\mu }\mathrm{s}$ per joint decision, whereas complete control-loop timing is dominated by network-statistics polling. These results support offline multi-agent SDN control as a competitive, low-overhead option when interpreted jointly with topology structure, flow completion, and strong heuristic baselines. Full article

Urban trees are important biological carbon pools in cities, yet their standing carbon stocks and the associated methodological uncertainty regarding their evaluation often vary among functional spaces. This study assessed 2442 individual trees representing 148 species in 116 standard 20 m × 20 m plots in the core urbanized area of Qingdao, China. Individual-tree carbon stocks were estimated from diameter at breast height, tree height, wood basic density, and allometric equations and were analyzed across urban functional units (UFUs), DBH classes, and species. Under the baseline Chave model, the sampled trees stored 153.337 Mg C, equivalent to 562.237 Mg CO₂, with a mean carbon density of 33.047 Mg C ha⁻¹. Residential districts contributed the highest total carbon stock, whereas public affairs service districts had the highest carbon density. Trees with DBH ≥ 40 cm accounted for only 3.93% of individuals but contributed 29.50% of the total carbon stock, and the top 20 species contributed approximately 79.0%. Alternative China-specific equations increased total estimates by 9.38–11.24% but covered only 10 of 148 species and 21.38% of trees at most; the UFU ranking and the main size- and species-contribution patterns remained unchanged. A root-to-shoot parameter sensitivity analysis likewise affected absolute estimates but not relative patterns. These findings support conserving mature trees, maintaining medium-sized trees as future carbon reserves, and targeting greening improvements to low-carbon-density functional spaces while recognizing the uncertainty inherent in absolute carbon-stock estimates. Full article

Wearable and intelligent bioelectronic systems are emerging as a key enabling technology for continuous, non-invasive health monitoring, coupling physiological sensing with data-driven inference. Within this paradigm, electroencephalography (EEG) provides a wearable-compatible biosensing modality for capturing the pre-sleep neurophysiological dynamics linked to emotional regulation and sleep onset. Insomnia affects approximately 10–15% of adults worldwide and is often associated with dysregulated emotions and pre-sleep hyperarousal. Existing EEG-based affective and sleep-onset processing pipelines often rely either on deep-learning architectures with limited interpretability or on hand-crafted spectral descriptors with weak theoretical motivation. This study presents an exploratory proof-of-principle bioelectronic processing framework in which EEG sensing features are organized according to ANT-7 (artificial neurotransmitter seven-dimensional model), a neurotransmitter-inspired computational taxonomy introduced as a heuristic feature-design prior rather than as a validated neurochemical theory. The proposed feature set includes the alpha/theta power ratio, sample entropy, Higuchi fractal dimension, and phase-locking value extracted from the public DREAMER and DEAP datasets (23 and 32 subjects, respectively). SVM, Random Forest, and 1D-CNN classifiers are trained under subject-independent leave-one-subject-out cross-validation with strict within-fold normalization to prevent data leakage, and interpretability is assessed through SHAP values and permutation importance (PI). To stress-test whether this feature organization transfers beyond the affective benchmarks on which it is trained, classifier outputs are then related to sleep-onset latency in Sleep-EDF Expanded through a deliberately cautious cross-dataset transfer analysis. Within this protocol, the best model reaches 88.4% accuracy in three-class affective-state recognition (stress/neutral/relaxed; AUC-ROC = 0.93). As an exploratory secondary analysis, classifier-derived relaxation estimates show a statistically significant negative association with polysomnographic sleep-onset latency and improve over a single alpha/theta-ratio baseline; this cross-dataset result is reported as a proof of concept, not as a validated sleep-onset predictor. Interpretability analyses (SHAP and permutation importance) indicate that the learned feature rankings are internally consistent with the neurotransmitter-inspired feature design, a property we interpret as internal coherence rather than as independent confirmation of the taxonomy. Together, these elements outline a complete sensor-to-AI processing chain—from EEG biosensing, through neurotransmitter-inspired signal-feature extraction, to interpretable and computationally lightweight inference—designed for compatibility with low-density wearable EEG devices and edge deployment. However, EEG does not measure neurotransmitter concentrations, the study does not benchmark ANT-7 directly against competing taxonomies such as valence-arousal/circumplex or RDoC-inspired feature organizations, and the Sleep-EDF analysis should not be interpreted as evidence that the model measures a validated latent construct of sleep readiness. Accordingly, the manuscript should be read as a framework-validation study of one interpretable feature taxonomy, not as a theory-validation study of ANT-7 or as a clinical validation study. Full article

The rapid expansion of make-on-demand and public chemical libraries has made exhaustive docking-based structure-based virtual screening increasingly difficult. This study introduces SurroDock, a lightweight deep-learning surrogate designed to approximate AutoDock Vina docking scores from low-cost two-dimensional molecular features, serving as a practical pre-filter for docking. SurroDock was evaluated for estrogen receptor alpha using two distinct conformations: an agonist-bound (PDB ID: 1GWR) and an antagonist/SERM-bound (PDB ID: 3ERT). The dataset comprised approximately 334,000 unique compounds curated from the NCI Open Database, PubChem, and BindingDB, all docked using a standardized AutoDock Vina workflow. The model was trained on concatenated 2D molecular representations comprising Morgan fingerprints, MACCS keys, RDKit physicochemical descriptors, Vina-inspired ligand descriptors, atom-pair fingerprints, and 2D pharmacophore fingerprints. The docking-score distributions differed substantially between receptor states, with 3ERT exhibiting more favorable scores than 1GWR and weak inter-state score correlation supporting state-specific modeling. Using the integrated Unified-200k training set (200,000 compounds randomly sampled per receptor from the three docked sources), SurroDock achieved strong held-out validation performance, with R² values of approximately 0.88 for 1GWR and 0.93 for 3ERT. In retrospective screening-style evaluation, SurroDock recovered substantial fractions of Vina’s top-ranked compounds at the top-1% recall (Recall@1%) of approximately 0.57 and 0.61 for 1GWR and 3ERT, respectively, yielding corresponding enrichment factors (EF@1%) of approximately 57-fold and 61-fold relative to random selection. Overall, the results indicate that 2D-based docking-score surrogate modeling can provide a reproducible and retrainable strategy for large-scale structure-based virtual screening by concentrating docking resources on a smaller, enriched subset of compounds. Because SurroDock emulates a docking scoring function rather than experimental binding affinity, its predictions should be used as prioritization aids and complemented by confirmatory docking, pose inspection, and experimental validation. Full article

We present a rank-based method for detecting short-duration gravitational-wave transients in 46 days of coincident data from the first Advanced LIGO observing run (O1). The method applies a moving-window implementation of Chatterjee’s rank correlation coefficient to whitened interferometric sensor strain data. This produces a computationally efficient statistic sensitive to temporally ordered signal structure without relying on waveform templates. Compared with traditional excess power and coherent burst searches, the rank-based formulation is potentially less sensitive to certain non-Gaussian noise transients. Furthermore, it processes dual-detector data faster than real time on a single CPU core. We evaluate the method using 60 hardware injections from the O1 dataset, recovering 28 compact binary coalescence injections, primarily for events with a single-detector signal-to-noise ratio above approximately 13. The pipeline identifies 31 transient candidates, including the astrophysical event GW150914 and two instrumental glitches. Although the present implementation is less sensitive than established search pipelines, these results demonstrate the feasibility of the new method. Rank-based detection statistics provide a computationally efficient and complementary method for low-latency transient detection in interferometric sensor networks. Full article

Soybean grain yield across extended planting calendars depends on sowing date, cultivar maturity group, and plant population. We evaluated BRS 1061 IPRO (MG 6.1) and DM 66i68 IPRO (MG 6.6) sown from September to January in 2021/2022 and 2022/2023 in Londrina, Brazil, at 200,000, 280,000, 360,000, and 440,000 viable seeds ha⁻¹ in a split-plot randomised complete block design. Grain yield, final stand, and yield components were analysed using mixed-model ANOVA, mean comparisons, regressions, correlations, principal component analysis, clustering, and descriptive permutation-based predictor ranking. Sowing date produced the largest performance gradient: grain yield ranged from 314 to 5771 kg ha⁻¹, and January sowings were consistently low yielding. Final stands ranged from approximately 1.6 × 10⁵ to 4.2 × 10⁵ plants ha⁻¹. Grain yield was strongly associated with grains m⁻² (r = 0.92), while PC1 and PC2 explained 71.3% of multivariate variation. Higher seeding rates did not provide a uniform advantage; intermediate rates were beneficial only in specific sowing-date × cultivar contexts, whereas 440,000 seeds ha⁻¹ sometimes reduced early-sowing yield. Seeding-rate decisions should, therefore, be cultivar- and date-specific, and January sowing is not recommended in this environment. Full article

Thermal runaway in electric vehicle power battery packs is a key risk in fire prevention and control for electric transportation. Its triggering, propagation, and failure modes are jointly affected by external thermal abuse, material insulation performance, and side reactions inside cells. To identify the temperature response and fire risk of power battery packs under different thermal abuse intensities, this study established a three-dimensional multiphysics thermal runaway simulation model in COMSOL Multiphysics 6.1, coupling solid heat transfer, electrochemical heat generation, and side-reaction heat release. A semi-quantitative risk ranking was then performed using failure mode, effects, and criticality analysis (FMECA). The model considered the low-temperature safe conditions, 120 °C, 140 °C, and 170 °C, as the main ambient temperature conditions, while also analyzing the effects of the heat transfer coefficient on trigger time and peak temperature. The results show that, under the low-temperature safe condition and the 120 °C condition, the battery module mainly exhibits slow heating and does not undergo thermal runaway. Based on the side-reaction characteristics, the temperature near 125 °C can be used as a risk warning threshold for thermal runaway. At 140 °C, the side-reaction heat source increases markedly, and the system enters the thermal runaway risk region. Because the trigger time is strongly affected by the heat transfer coefficient and monitoring position, this condition is interpreted only as a risk-acceleration stage under critical thermal abuse. Approximately 167 °C can be regarded as the critical threshold for irreversible thermal runaway. Under severe thermal abuse at 170 °C, rapid intensification of internal side reactions increases the peak module temperature to 375–385 °C. Temperature field evolution shows that heat is transferred mainly from the exterior to the interior before thermal runaway, forming an outside-high- and inside-low-temperature distribution. After the runaway stage begins, heat release from internal cell side reactions becomes dominant, and the high-temperature region concentrates inside the module, producing a gradient reversal with a higher internal temperature. The FMECA results show that the positive electrode–electrolyte reaction has the highest RPN, with a value of 405. Accelerated SEI decomposition and the negative electrode–electrolyte reaction also form key risk links in the chain heat-release pathway. This study provides a reference for thermal management, fire barrier design, and fire risk classification of power battery packs. Full article

The continuous emergence of new attack classes challenges intrusion detection in edge Internet of Things (IoT) networks. Although federated learning enables distributed devices to collaboratively train a shared detector without exchanging raw traffic data, most federated intrusion detection systems assume a fixed label space. Retraining with all historical data incurs substantial storage and computation costs, whereas updating only with newly collected samples can cause catastrophic forgetting. The detector must mitigate catastrophic forgetting of previously observed attack classes while preserving sufficient new-class plasticity to learn emerging attacks under highly non-IID device data, intermittent client availability, constrained local memory, and repeated communication over bandwidth-limited and intermittently connected links. To address these challenges, this paper proposes EdgeFedCIL, a communication-efficient federated class-incremental intrusion detection framework. EdgeFedCIL preserves historical knowledge through client-local replay and knowledge distillation while reducing repeated model transmission through adaptive low-rank compression, quantization, and error feedback. A classifier-head protection strategy further limits compression-induced degradation of class discrimination. Experiments on public intrusion-detection datasets show that EdgeFedCIL achieves competitive or superior detection and historical-knowledge retention performance, particularly under highly heterogeneous client distributions, while reducing cumulative client-to-server model transmission by up to approximately 10.54 times relative to full-precision transmission. These results demonstrate the effectiveness of EdgeFedCIL for continual and communication-efficient intrusion detection in resource-constrained edge IoT networks. Full article

Background: Surgical sepsis complicated by acute kidney injury (AKI) is associated with high mortality, and early risk stratification remains challenging. The prognostic nutritional index (PNI), derived from serum albumin and lymphocyte count, reflects nutritional and immune status, but its prognostic value in surgical sepsis with AKI has not been well defined. Methods: In this retrospective cohort study based on the Medical Information Mart for Intensive Care IV (MIMIC-IV) (version 3.1) database, we included adult patients admitted to a surgical or surgery-related intensive care unit (ICU) between 2008 and 2022 who met the Sepsis-3 criteria and developed AKI according to KDIGO. The primary endpoint was 90-day all-cause mortality. We used multivariable Cox proportional hazards models, restricted cubic splines, Kaplan–Meier analysis, and predefined subgroup analyses to examine the association between PNI at ICU admission and 90-day mortality. Results: A total of 1483 patients were included, with a 90-day mortality rate of 30.2%. Non-survivors had a lower median PNI than survivors (35 vs. 36, p < 0.001). After sequential adjustment for demographics, comorbidities, illness severity, and major interventions, each 1-point increase in PNI was associated with a 1.5% reduction in 90-day mortality (hazard ratio 0.985, 95% confidence interval 0.974–0.997, p = 0.012). Restricted cubic spline analysis showed an approximately linear inverse relationship between PNI and mortality risk (p for overall association = 0.046; p for nonlinearity = 0.534). Using an optimal cut-off of 29.51, patients with low PNI had significantly lower 90-day survival than those with high PNI (log-rank p < 0.0001), and subgroup analyses demonstrated generally consistent protective associations across age, sex, illness severity, and key treatments. In propensity score-matched analysis, the association was attenuated and no longer significant (HR = 1.174, 95% CI: 0.901–1.529, p = 0.234), although the direction of effect remained consistent. Conclusions: Lower PNI at ICU admission was associated with higher 90-day mortality in multivariable-adjusted models and may serve as a simple, routinely available adjunctive marker for early risk stratification in surgical sepsis patients with AKI. However, given the non-significant finding in propensity score-matched analysis, its independent prognostic value remains uncertain, and further prospective studies are needed to validate its clinical utility. Full article

Burned-area totals are useful for national monitoring, but they do not reveal how, when or under what social and ecological conditions a landscape burns. We developed an event-based fire-regime and stewardship framework for Eswatini, a topographically compressed southern African country where protected areas, communal rangelands, cropland margins, plantation landscapes and peri-urban interfaces occur in close proximity. Global Fire Atlas event histories for 2001–2025 were organised by fire year and intersected with approximately 10 km² hexagonal units. The burned-area rate, event frequency, recurrence, seasonality, large-fire dominance, pyrodiversity and trend were used to classify fire-regime types independently of socio-ecological predictors. An XGBoost regression model, evaluated on a 20% held-out test set, was interpreted using exact TreeSHAP diagnostics. Fire activity was strongly seasonal: July–September accounted for 78.2% of the burned area, with August alone accounting for 34.2%. Eight fire-regime types were identified, ranging from low-information and episodic units to frequent small-fire mosaics, large-fire-dominated areas and emerging burned-area intensification regimes. The burned-area-rate model performed well on held-out data (R2 = 0.71; Spearman rho = 0.75). Human modification, goat density, elevation, forest probability, fuelwood dependence and precipitation seasonality ranked among the most influential predictors, but their fitted effects were non-linear and often bidirectional. The combined diagnostics supported six adaptive management zones covering protected-area stewardship, conservation-sensitive management, settlement–livelihood interfaces, late-season risk reduction, monitoring and integrated landscape management. Although the Eswatini results are context-specific, the workflow offers a transferable way to connect fire histories, socio-ecological contexts and place-based stewardship in African mosaic landscapes. Full article

This paper develops an adjoint-based framework for jointly reconstructing a space–time-dependent internal heat source and an unknown initial temperature field from sparse, noisy measurements. A single adjoint solve supplies the gradients with respect to both fields, so each iteration requires one forward and one adjoint solve. Deterministically, CGLS with discrepancy-principle stopping reaches a comparable regularized solution in about an order of magnitude fewer iterations than Landweber–Fridman. In the Bayesian formulation, Gaussian noise and Matérn priors yield an exact Gaussian posterior; prior-preconditioned conjugate gradients compute the maximum a posteriori estimate, while a low-rank approximation of the prior-preconditioned data-misfit Hessian provides pointwise credible bands. The exact discrete adjoint gives machine-precision gradients, and prior-predictive experiments verify nominal pointwise coverage. Numerical experiments compare the reconstructions and assess sensitivity to noise, discretization, and prior hyperparameters. The Bayesian reconstruction is more accurate and mesh-robust in the reported tests. A calibrated generalized-${\chi }^2$ discrepancy diagnostic detects misspecification caused by an omitted initial-temperature offset and, less strongly, by discontinuous sources outside the prior model. These experiments demonstrate joint reconstruction and scalable uncertainty quantification using only forward and adjoint heat-equation solves. Full article