Search Results (9,584)

Accurate traffic flow prediction remains challenging due to the complex spatio-temporal dependencies inherent in road networks. Existing graph-based prediction models typically construct the graph structure from geographical adjacency. This limits spatial feature extraction to physically neighboring nodes. Consequently, potential correlations among non-adjacent nodes within the same region are overlooked. Moreover, these models generally treat the influence of neighboring nodes uniformly, whereas, in practice, such influences are inherently heterogeneous. On the temporal side, many approaches rely on recurrent units such as Long Short-Term Memory (LSTM) or Gated Recurrent Units (GRUs) to capture short-term dependencies, yet these architectures are prone to gradient vanishing or explosion, leading to training instability. To address these issues, this paper proposes Seq2-ResGCRN, an encoder–decoder framework based on residual graph convolutional recurrent networks with an attention mechanism. The model reconstructs the adjacency matrix via graph diffusion convolution to learn adaptive edge weights, thereby capturing heterogeneous spatial influences. Seq2-ResGCRN further integrates graph convolution into the GRU architecture and introduces residual connections to alleviate gradient degradation during training. This study conducts experiments on two open-source real-world datasets (i.e., PEMS04 and PEMS08) to evaluate the Seq2-ResGCRN. The raw data, originally collected at a sampling frequency of 30 s, are aggregated into 5 min time intervals. Each record comprises three features: traffic flow, traffic speed, and road occupancy. The experimental results demonstrate that our Seq2-ResGCRN outperforms state-of-the-art methods, achieving 1.6–2.3% and 2.1–5.5% relative improvements in MAE and RMSE, respectively. Seq2-ResGCRN effectively captures the spatio-temporal correlations of short-term traffic flow and achieves superior predictive performance. Full article

We propose a decentralized multi-agent reinforcement learning (MARL) framework enhanced with graph neural networks (GNNs) for dynamic scheduling optimization in sports buildings, addressing the limitations of conventional centralized approaches. The system models each facility as an autonomous agent that learns adaptive policies through proximal policy optimization, where local observations include real-time occupancy, energy consumption, and user preferences. Global coordination is achieved via a GNN-based communication protocol, which captures spatial–temporal dependencies between facilities by treating the building as a graph with nodes representing facilities and edges encoding connectivity. The framework incorporates prioritized experience replay to handle unexpected events such as equipment failures or demand spikes, ensuring robust adaptation in simulated real-world scenarios. Moreover, the reward function dynamically balances occupancy utilization, energy efficiency, and user satisfaction with adaptive weights adjusted through meta-learning. The proposed method interfaces with existing facility management systems by translating agent actions into control signals for HVAC, lighting, and staff scheduling. Implemented using Vision-Transformer hybrids and GraphSAGE architectures, the framework supports federated learning to preserve data privacy across facilities. Experimental results in simulation demonstrate significant improvements, suggesting strong potential for deployment in large-scale, heterogeneous environments. This work advances the state-of-the-art in intelligent building management by enabling decentralized decision-making while maintaining global coordination through graph-structured communication. Full article

Accurate Land Use and Land Cover (LULC) mapping in high-resolution alpine environments is challenging due to complex terrain, heterogeneous vegetation, seasonal snow and ice cover, and the limited spectral information provided by conventional aerial imagery. Although foundation models such as the Segment Anything Model (SAM) effectively capture structural features, their class-agnostic design, limits fine-grained semantic discrimination and typically requires large annotated datasets. This study proposes a multi-modal deep learning framework for alpine LULC mapping using sparse annotations, which would fall under the category of weakly supervised learning. The framework employs a dual-encoder architecture that integrates RGB imagery, six-band multispectral imagery, and custom adapters for spectral indices, and Digital Surface Models (DSMs). A SAM-based encoder extracts geometric and contextual features from RGB imagery, while a dedicated encoder learns complementary spectral representations from multispectral data. To address the boundary uncertainty introduced by sparse supervision, we propose post inference hybrid refinement strategy that combines a Canopy Height Model (CHM) derived from the DSM to improve tree crown delineation with edge-based refinement for low vegetation classes, such as shrubs, and mathematical methods to refine road and building edge delineation with DSMs. Experimental results across fourteen alpine classes highlight that the framework achieves a validation mIoU of 0.777, macro-averaged over the fourteen classes. For the present sensor choice and classification scheme, no comparable multi-modal baseline exists. Thus, results are reported in absolute terms. The present multi-modal data fusion sets the baseline for future scalable alpine LULC mapping applications. Full article

Tracked agricultural robots operating on soft ground may experience asymmetric left–right track slip, while two-dimensional global navigation satellite system (GNSS) observations may contain a dominant anomaly in one coordinate. This simulation study evaluates a robust multi-sensor localization method for this coupled degradation. Independent left–right slip ratios are introduced into a six-state extended Kalman filter (EKF) to improve motion prediction. A coordinate-adaptive Huber extended Kalman filter (CAH-EKF) is then developed for GNSS updating. It combines component-wise normalized innovations and innovation jumps for coordinate-level anomaly detection, maps Huber weights to effective measurement covariance, and uses finite-state switching to confirm, maintain, and recover robust processing while retaining information from the relatively reliable coordinate. Simulation experiments on S-shaped and multi-U paths separate prediction-stage and update-stage effects. The independent-slip model improves over the common-slip model by 16.99–34.34% in position root mean square error (RMSE) and 29.11–36.27% in heading RMSE. Under sustained single-coordinate GNSS anomalies, CAH-EKF outperforms IS-EKF and remains competitive with Huber-IEKF. Under the main joint degradation experiments, the combined method yields 42.78–77.43% position and 27.05–76.33% heading improvements. Additional simulation checks covering slip excitation, path and motion changes, GNSS-degradation variants, and runtime further characterize the operating range of the method. These results support robust low-cost localization using encoders, a yaw-rate gyroscope, and a two-dimensional GNSS within the tested simulation conditions. Full article

Background/Objectives: Parkinson’s disease (PD) is associated with gait abnormalities that provide quantitative information on motor function. This study developed and evaluated a signal-to-image deep learning framework for distinguishing participants with diagnosed PD from healthy controls using plantar pressure gait signals. Methods: VGRF signals from the PhysioNet Gait in Parkinson’s Disease Database were transformed into RGB images encoding left-foot pressure, right-foot pressure, and the absolute bilateral difference. ResNet50, EfficientNet-B0, ViT-Tiny, and Swin-Tiny were evaluated at three resolutions and compared with 1D-CNN and BiLSTM baselines. Repeated subject-level five-fold cross-validation with three repetitions and participant-level bootstrap analysis were performed. Results: Among RGB models, ViT-Tiny at 384 × 384 achieved a mean accuracy of 77.64%, F1-score of 82.82%, and AUC of 86.40%. In the Swin-Tiny 384 × 384 ablation, the absolute bilateral-difference representation achieved the highest mean AUC (86.68%). However, its participant-level AUC advantage over RGB was not statistically conclusive (ΔAUC = 0.030, 95% CI: −0.011 to 0.068). Conclusions: Signal-derived gait images provide a feasible approach for PD classification, with bilateral-difference information showing potential discriminative value. Full article

A perfect Italian dominating function assigns values in $\{0,1,2\}$ to the vertices of a graph so that every zero-valued vertex has a neighbor-value sum of exactly two. We determine the computational complexity of the problem for bipartite graphs whose neighborhoods are subtrees of a simple host tree. On convex bipartite graphs, whose host is a path, an explicit ten-coordinate boundary dynamic programming algorithm computes the optimum value in $O\left(n^{11}\right)$ time and $O\left(n^{10}\right)$ space. We extend the method to the annotated setting in which the input supplies a size-k vertex set whose deletion leaves a convex bipartite graph, together with a convex representation of the remainder. Enumerating the modulator labels and tracking four-valued exact-sum counters gives an $O\left(k{6}^k{n}^{11}\right)$-time fixed-parameter tractable algorithm. On triad-convex bipartite graphs, whose host consists of three paths with a common endpoint, we give a polynomial-time algorithm by combining a constructible mim-width-three decomposition with a three-stable locally checkable formulation. More generally, the same method is polynomial for $(t,\Delta )$-tree-convex bipartite graphs for every fixed pair $(t,\Delta )$, provided that the representation is supplied. In contrast, we prove that the decision problem is NP-complete on star-convex bipartite graphs. The reduction from Restricted Exact Cover by 3-Sets uses a forced label-2 center and constant-size set gadgets whose two minimum-weight modes encode an exact cover. We further prove that the problem parameterized by the target weight is W[1]-hard on the same graph class through a parameter-preserving reduction from Perfect Code. Full article

In open, noncooperative electromagnetic environments, prior information about target signals, such as their modulation types and symbol rates, is often difficult to obtain. Consequently, both conventional methods and explicitly label-dependent jamming waveform generation methods face substantial limitations in practical applications. To address this issue, we propose a Multiscale Aggregation Generative Adversarial Network with Masked Pre-training (MA-GAN), a communication jamming waveform generation method based on masked self-supervised learning and multiscale feature fusion. We construct a masked waveform modeling pre-training framework for In-phase/Quadrature (I/Q) time-series signals, through which the Masked Waveform Modeling encoder (MWM-encoder) learns hierarchical and multiscale representations from unlabeled data. The generator is guided by multiscale statistical features and combined with a multi-head multiscale discrimination mechanism and physical constraints to improve the structural consistency between the generated jamming waveforms and the target signals. Experimental results show that, under Additive White Gaussian Noise (AWGN), Rayleigh fading and Rician fading channels and across different Signal-to-Noise Ratio (SNR) and Jamming-to-Signal power Ratio (JSR) conditions, MA-GAN achieves effective jamming performance for five modulation types observed during training and five unseen modulation types, with performance approaching that of the Signal-Like method, which assumes complete prior information. These results indicate that MA-GAN can generate communication jamming waveforms with a certain degree of cross-modulation and robustness to the tested channel conditions without relying on explicit prior information. Full article

Automatic modulation recognition (AMR) is a key enabling technique for intelligent spectrum sensing, non-cooperative wireless signal analysis, and communication monitoring. However, its reliability degrades significantly under low signal-to-noise ratio (SNR) conditions, where noise weakens local I/Q waveform patterns, blurs time-frequency structures, and limits the robustness of single-domain representations. Although multimodal AMR methods have improved recognition by combining I/Q and time-frequency representations, many of them still use fixed fusion rules, which may not reflect the changing reliability of different modalities across SNR levels. In addition, most existing backbones model signal sequences in a Euclidean manner and lack an explicit mechanism to capture adaptive relational structures among signal tokens under strong noise. To address these limitations, this paper proposes MPCG-Net, a modulation-prior-enhanced cross-modal graph learning framework for low-SNR AMR. MPCG-Net first performs modulation-prior-enhanced cross-modal encoding, where learnable multi-resolution spectro-temporal representation learning enhances the input representation and training-stage SNR-aware cross-modal consistency alignment improves the complementary interaction between I/Q and spectro-temporal tokens. A modulation context memory module then refines the fused tokens before graph construction. In the graph-based relational learning stage, a temporal similarity graph with self-loops, local temporal-chain edges, and top-k feature-similarity edges is built to learn more discriminative graph-level signal representations. Experiments on RadioML2016.10A, RadioML2016.10B, and RML22 validate the effectiveness of the proposed design. Compared with the strongest baseline on each dataset, MPCG-Net improves the average accuracy in the $-20$ to 0 dB low-SNR range by 6.90, 6.45, and 1.66 percentage points on RadioML2016.10A, RadioML2016.10B, and RML22, respectively, while maintaining a lightweight model scale. Full article

Vibration-based diagnosis can identify transformer core looseness non-intrusively, but subtle vibration differences and noise-disturbed Gramian angular field (GAF) textures reduce diagnostic accuracy at low signal-to-noise ratios (SNRs). This paper proposes a variational mode decomposition (VMD)-assisted dual-representation cross-interaction network (DGCM-Net) with multi-SNR training. VMD parameters are optimized by particle swarm optimization (PSO), and signals reconstructed from selected modes are encoded as paired Gramian angular summation field (GASF) and Gramian angular difference field (GADF) images. Two independent DenseNet121 branches extract features from the two representations. CrossGAF enables cross-branch interaction before fusion, while the residual bottleneck multi-scale selective kernel (RBMSSK) module processes fused features at different scales. Across five independent runs, the proposed method achieves 99.84 ± 0.11% accuracy under the standard condition and 97.39 ± 0.23% mean accuracy across four noisy test sets, with 91.32 ± 0.62% accuracy at 10 dB. On a public 50 kVA transformer dataset, retraining yields 98.20 ± 0.65% accuracy at 2.5 dB. In cross-transformer few-shot adaptation, using 5% of the target-domain training samples increases the mean accuracy across four noisy conditions from 93.63% to 97.16% over target-only training. Full article

1-deoxy-D-xylulose-5-phosphate synthase (DXS) is the first key enzyme in the methylerythritol phosphate (MEP) pathway of plant terpenoid biosynthesis, and it plays a vital role in Eucommia ulmoides terpenoid biosynthesis. In this study, bioinformatics methods were used to comprehensively identify and analyze the expression pattern of the EuDXS gene family, aiming to provide a basis for further functional study of EuDXS genes. A total of four EuDXS gene family members were identified and named EuDXS1 to EuDXS4. The encoded proteins contained 625 to 713 amino acid residues, with molecular weight ranging from 67.98 kDa to 76.61 kDa. The theoretical isoelectric points varied from 6.79 to 8.69, aliphatic index was between 85.29 and 91.29. In silico subcellular localization prediction revealed that all EuDXS proteins were localized in chloroplasts. EuDXS gene members were categorized into three subfamilies, which were unevenly distributed on three chromosomes. The promoters of EuDXS genes contained various cis-acting elements related to stress response, phytohormone signaling, light response and growth regulation. Expression pattern analysis showed that EuDXS genes exhibited tissue-specific expression: EuDXS1 was highly expressed in stem, leaf and fruit, EuDXS2 was predominantly expressed in fruit. EuDXS1 and EuDXS2 exhibited high expression levels at the early developmental stage of fruits and leaves. In addition, EuDXS genes responded to salt and drought stress in varying degrees. Transient expression in tobacco revealed that EuDXS1 and EuDXS2 could increase carotenoid and total chlorophyll content. This study will provide important genetic resources for further exploration of EuDXS gene function and germplasm innovation in E. ulmoides. Full article

Background/Objectives: Macrophages are key contributors to the pathogenesis of rheumatoid arthritis (RA) and osteoarthritis (OA), but their disease-associated immunometabolic states remain incompletely understood. Methods: Here, we comparatively analyzed patient-derived transcriptomic data from synovial tissues and synovial macrophages in RA and OA. Results: RA synovial tissues showed stronger interferon-γ (IFNγ)-signatures and reduced mitochondrial Complex I-related programs compared with OA. Similar reductions in oxidative phosphorylation, mitochondrial translation, and Complex I-related transcription were observed in synovial fluid-derived RA macrophages relative to control macrophages. In an independent dataset of synovial tissue-derived OA macrophages, inflammatory-like OA macrophages exhibited concurrent enrichment of glycolytic and oxidative metabolic programs. Bulk RNA-seq analysis of synovial monocytes within a common study framework further showed that reduced mitochondrial translation and Complex I-related programs were preferentially associated with leukocyte-rich inflammatory RA. At single-cell resolution, IFN-responsive monocytes enriched in leukocyte-rich RA showed reduced transcriptional representation of both nuclear- and mtDNA-encoded Complex I components. Conclusions: Collectively, these findings link an IFN-responsive inflammatory myeloid state in RA with coordinated suppression of mitochondrial translation and Complex I-related transcription, suggesting a transcriptional state that may increase susceptibility to mitochondrial Complex I dysfunction during persistent inflammation. Full article

Background/Objectives: Glioblastoma surveillance after surgery and chemoradiation remains challenging because MRI findings of tumor recurrence can overlap with pseudo-progression, treatment-related effects, and postoperative tissue changes. Methods: We developed a seven-channel postoperative MRI framework using a deep learning transfer method to support non-invasive glioblastoma treatment-effect assessment. The model used T1 contrast-enhanced, FLAIR, T1, RSI-Cell, ADC, T2, and cerebral blood flow volumes from 124 postoperative glioblastoma patients (164 MRI timepoints) as input. Images were processed using a 3D ResNet18 encoder pretrained on 588 postoperative glioma samples and fine-tuned using task-specific classification heads. The cohort comprised 124 patients contributing 164 postoperative MRI timepoints, all acquired at 3T on scanners from a single vendor. Performance was evaluated with nested five-fold cross-validation stratified and assigned at the patient level, so that all timepoints from a given patient fell in one-fold and the training epoch was selected on an inner split rather than on the fold being reported. Because some of the clinical labels were incomplete, the number of evaluable timepoints differed by task (recurrence versus pseudo-progression, 164; MGMT, 99; short-term survival, 139). The whole procedure was repeated under three independent random seeds and results are reported as the mean and standard deviation across seeds. Results: The strongest clinical endpoint was recurrence versus pseudo-progression, where nested cross-validation across three random seeds gave a pooled out-of-fold AUC of 0.935 (SD = 0.014), area under the precision-recall curve of 0.973, balanced accuracy of 0.880, sensitivity of 0.917, and specificity of 0.843. No other endpoint reached reliable discrimination. Radiation decision reached an AUC of 0.658 (SD = 0.039), while MGMT promoter methylation (AUC = 0.532, SD = 0.082) and short-term survival (AUC = 0.514, SD = 0.027) were indistinguishable from chance. Conclusions: In this single-center proof-of-concept study, postoperative MRI successfully distinguished tumor recurrence from pseudo-progression. However, the models did not reliably predict the other three outcomes related to molecular status, treatment planning, and prognosis. Overall, the model learned imaging features specifically associated with recurrence, rather than a more general representation of the tumor that can predict many different clinical outcomes. These findings support technical feasibility for a single endpoint rather than clinical readiness, for which external multi-center validation is required. Full article

Deep learning has achieved notable success in sensor-based human activity recognition (SHAR), yet wearable inertial signals contain activity patterns at different temporal scales while practical deployment imposes strict computational constraints. This paper proposes a three-stage Adaptive CNN-Enhanced Scale-fusion Network (ACESNet) for six-channel accelerometer–gyroscope activity recognition. The Adaptive Kernel Encoder dynamically combines temporal kernels of different sizes, while the CNN-Enhanced Scale-fusion design combines multi-scale temporal mixing, explicit channel mixing and a parallel local CNN path without materializing a global attention matrix. The main benchmark uses a fixed stratified 70/15/15 window-level partition (split seed 42), and all methods in the main comparison are repeated over five training seeds. On REALWORLD, MotionSense, UCI-HAR and SHL, ACESNet achieves accuracies of 95.37 ± 0.09%, 99.33 ± 0.12%, 98.01 ± 0.23% and 93.49 ± 0.10%, with Macro-F1 scores of 95.59 ± 0.09%, 99.12 ± 0.15%, 98.16 ± 0.21% and 94.03 ± 0.08%, respectively. A separate official subject-independent UCI-HAR evaluation yields 94.01 ± 0.84% accuracy and 94.11 ± 0.88% Macro-F1. The adaptive-kernel selector has a mean normalized entropy of 0.9114, providing evidence against severe single-branch collapse rather than strong kernel selectivity. On a Raspberry Pi 5 (Raspberry Pi Ltd., Cambridge, UK), ACESNet requires 8.731 MFLOPs and 1.606 ± 0.014 ms per window in FP32 ONNX Runtime 1.18.1. These results support a favorable performance–efficiency trade-off relative to the selected baselines under the stated protocol. Full article

The growing scale of photovoltaic (PV) installations creates an urgent need for automated fault detection systems that operate in real time on resource-constrained monitoring hardware. Although deep learning methods achieve high classification accuracy on PV fault benchmarks, their computational requirements make them impractical for deployment on embedded devices such as smart inverters and industrial IoT controllers. This paper proposes a lightweight classification approach based on the fuzzy k-Nearest Neighbor (Fuzzy kNN) algorithm, in which every electrical measurement is represented as an Ordered Fuzzy Number (OFN) whose spread is automatically calibrated from the local standard deviation of the measurement window. This adaptive fuzzification encodes the inherent sensor noise and environmental variability of SCADA measurements without any manual parameter tuning. Six fuzzy distance metrics, obtained by combining three defuzzification operators (FOM, LOM, MOM) with the Euclidean and Manhattan distance functions, were evaluated on the public GPVS-Faults benchmark containing approximately 1.8 million samples describing seven fault types in a grid-connected PV system operating under MPPT and IPPT control. In binary anomaly detection, the proposed method achieved an accuracy of 92.97% ± 1.17% (k = 3, MOM defuzzification with Manhattan distance), which is statistically comparable to the Random Forest baseline (92.42% ± 0.65%) while offering approximately one hundred times faster inference (1.2 ms versus 120 ms per sample) and a model footprint of only 2 MB. In multiclass fault type classification, the method reached 96.18% ± 1.80% accuracy against 97.88% ± 0.69% for Random Forest. A consistent and previously unreported observation is that the Manhattan distance systematically outperforms the Euclidean distance on three-phase electrical measurements, improving accuracy by approximately 0.90 percentage points across all tested configurations. The complete source code and the experimental pipeline are publicly released to ensure full reproducibility of the reported results. To assess generalization rigorously, a stratified group cross-validation was additionally performed in which all windows from a given experimental recording are confined to a single fold; under this leakage-free protocol every evaluated method, including Random Forest, degrades to the 50–62% range, which shows that cross-recording transfer is an intrinsic difficulty of the single-run GPVS-Faults benchmark rather than a weakness specific to the proposed classifier. Full article

Unsupervised feature selection is essential for high-dimensional data analysis, where irrelevant and redundant variables may obscure the intrinsic data structure, reduce interpretability, and degrade downstream learning performance. The key challenge is to identify informative features without label supervision while suppressing redundant selections under nonlinear dependencies. To address this issue, this paper proposes an unsupervised Hilbert–Schmidt Independence Criterion (HSIC)–Elastic Net (ENet) feature selection framework, termed U-HSIC-ENet. The proposed method reformulates unlabeled feature weighting as a self-supervised kernel alignment problem. Specifically, a target kernel is constructed directly from unlabeled data to encode global sample relationships, while each feature is represented by a centered and Frobenius-normalized feature-induced kernel. Feature relevance is then measured by centered kernel alignment (CKA) in a reproducing kernel Hilbert space (RKHS). The main novelty of U-HSIC-ENet lies in a unified relevance–redundancy–stability formulation. Feature–target alignment is used to estimate nonlinear relevance, whereas pairwise similarities between feature-induced kernels are used to characterize inter-feature redundancy in the same kernel alignment space. On this basis, an explicit off-diagonal redundancy penalty is incorporated into a nonnegative Elastic Net-type objective, which strengthens the suppression of co-selected similar features while preserving sparse and stable feature weighting. The resulting quadratic formulation clarifies how relevance promotion, redundancy control, sparsity, and numerical stabilization are coupled within a single optimization framework. Experiments on eight benchmark datasets under a fixed-budget evaluation protocol show that U-HSIC-ENet achieves the strongest average performance on Normalized Mutual Information (NMI), the Adjusted Rand Index (ARI), and clustering accuracy (ACC) compared with representative graph-, spectral-, and HSIC-based baselines. The advantage is the most pronounced on NMI, suggesting that the self-supervised target kernel and CKA-based relevance modeling are effective in preserving the clustering-relevant nonlinear structure. Friedman tests and Wilcoxon signed-rank tests with Holm correction provide statistical support for the observed improvements. Subsampling-based stability evaluation reveals a trade-off between clustering effectiveness and selection reproducibility: several baselines achieve higher stability scores despite the stronger average clustering performance of U-HSIC-ENet. These results indicate that the proposed framework is effective for unsupervised nonlinear feature weighting when relevance estimation, redundancy control, and stability are considered jointly. Full article