Search Results (181)

In the complex problem of air-to-ground mission planning, multi-UAV systems face significant challenges such as system complexity and heterogeneity, insufficient target observability, and difficulties in collaborating information sharing. To address these issues, this paper proposes a novel learnable sparse attention graph architecture (SAGA). This architecture deeply integrates graph reasoning and policy optimization within the MAPPO framework and includes three innovative mechanisms: (i) a GATv2-based graph neural network encoder that performs multi-round distributed consensus on the communication graph among UAVs via a multi-head attention mechanism, enabling selective aggregation of tactical information; (ii) an edge predictor that learns to prune low-value communication links, generating a sparse and mission-adaptive communication topology; and (iii) an L1 sparsity penalty term that further enhances communication efficiency. In a self-developed simulation environment for heterogeneous multi-UAV mission planning, comprehensive comparative experiments were conducted against the following baseline reinforcement learning algorithms: MADDPG, MATD3, QMIX, MAPPO, TarMAC, DGN, and G2ANet. The experimental results show that SAGA achieves reward values of 390 and 1100 in small-scale and large-scale scenarios, and outperforms the best-performing baseline algorithm by more than 20% across all operational performance metrics. Generalization experiments validate the model’s robust transfer capability under unknown defense deployment modes. Ablation experiments further confirmed the individual contributions of the three components. This study provides an innovative and effective method for mission planning of heterogeneous multi-UAV systems in partially observable adversarial environments. Full article

Satellite video provides repeated observations of the same ground scene within short acquisition intervals, but blur, detector sampling, radiometric differences, noise, and registration errors complicate joint reconstruction. This study presents mixed sparse representation-based collaborative quality improvement (MSR-CQI), a model-based method for joint radiometric and spatial reconstruction of short optical satellite video sequences. The method combines multiframe fidelity, effective PSF modeling, an intensity prior, overlapping group sparsity, high-order nonconvex regularization, intensity bounds, and optional static observation weighting. In controlled ×2 experiments with known HR references, MSR-CQI achieved 42.8714 dB PSNR and 0.9756 SSIM. With the same seven input frames, it achieved 43.1049 dB/0.9714, compared with 42.1369 dB/0.9698 for PnP-NLM and 38.1675 dB/0.9404 for DUF-16L. Retaining measured sampling shifts in the observation operators yielded 45.7461 dB/0.98046, versus 45.0380 dB/0.97904 after LR registration and resampling. The proxy derived from the reserved real frames instead favored the common-grid reconstruction, showing that agreement with this proxy does not establish recovery beyond the native sensor resolution. Static observation weighting was also scene-dependent. These results support retaining sampling phases explicitly in controlled spatial SR, while the real-data results after common-grid resampling are interpreted as multiframe restoration and proxy agreement. Full article

This paper proposes a post-training structured channel pruning scheme leveraging Gradient-Weighted Class Activation Mapping (Grad-CAM) for few-class fire detection under limited computational resources. After standard training, category-specific gradient signals extracted from the detection heads are used to generate channel-wise class activation maps over multi-scale feature layers, and fire and smoke responses are fused to rank channel importance. A layer-level retention quota is further applied to implement structured channel pruning, followed by lightweight fine-tuning. The pipeline does not require additional sparsity-inducing training. We validate the method on a self-established fire and smoke dataset containing 9041 images, using YOLOv5s, YOLOv5m, and YOLOv5l as baseline detectors. In workflow-level comparisons, the proposed method achieved higher mAP@0.5 than the implemented L1-based workflow at 40% and 60% pruning, but not at 80%. At 60% pruning, the pruned YOLOv5s model contained 2.158 M parameters and required 2.901 GFLOPs. These results indicate that Grad-CAM provides a useful class-aware criterion for channel importance and offers a promising model-compression strategy for resource-constrained fire detection, although physical edge-device performance remains to be evaluated. Full article

This paper proposes a novel Tensor Train (TT)-based tensor-on-tensor regression optimization framework for variable selection based on mode-1 hyperslice sparsity. The formulation incorporates an ${\mathcal{l}}_{2,0}$-regularized term on the first TT-core while imposing Stiefel manifold constraints on the remaining $M-1$ TT-cores. Leveraging the property that the group sparsity of the first core is equivalent to the hyperslice sparsity of the global structure, we establish theoretical guarantees for the uniform variable-selection consistency of the proposed model. To efficiently solve the proposed model, we design an alternating iterative algorithm equipped with a preconditioned metric and prove its convergence to a critical point. Extensive numerical experiments on both synthetic and real-world datasets demonstrate that the numerical solutions generated by our algorithm exhibit exact support recovery in practice, tightly aligning with our theoretical analysis. Full article

Graph-based nonnegative matrix factorization (GNMF) has been widely used for dimensionality reduction and data clustering because it can preserve the intrinsic geometric structure of data. However, many existing GNMF-based methods still rely on full sample similarity graphs, resulting in high computational costs; moreover, their sparsity constraints usually treat all features uniformly, making it difficult to distinguish structurally important features from redundant or noisy ones. To address these issues, this paper proposes a feature-graph-guided adaptive Log-L_2,1 sparse NMF with anchor dual graphs under a logarithmic framework. Specifically, anchor-based representations are simultaneously constructed in the sample and feature spaces to approximate the corresponding full-scale graphs. The sample anchor graph preserves the local manifold structure among samples, whereas the feature anchor graph plays a dual role: it preserves structural relationships among features and provides degree information for generating the adaptive weights g_i of the row-wise Log-L_2,1 penalty imposed on the basis matrix U. Consequently, structurally well-connected features receive weaker sparsity penalties, while weakly connected and potentially redundant features are more strongly suppressed. In addition, a logarithmic reconstruction framework is introduced to reduce the influence of large residuals caused by noise and outliers. These mechanisms jointly integrate sample structure preservation, feature structure preservation, and feature-aware sparse learning within a unified graph-NMF model. To optimize the model, multiplicative update rules are derived, while the nonnegativity of the factor matrices is inherently preserved throughout the iterations. Extensive evaluations on several benchmark datasets demonstrate the effectiveness and robustness of the proposed method. Full article

This research introduces a new spectral pruning approach using the Adjacency Edge Index (AEI), which is a centrality measure from the theory of spectral graphs and first-order matrix perturbation theory. The AEI score reflects the contribution of each neuron to the network’s dynamic synchronizability via the Fiedler vector. The AEI approach thus offers a mathematically motivated saliency score in the context of data-driven neuron co-activation graphs. The proposed approach has been tested on MNIST, Fashion MNIST, KMNIST, and a real-world social network dataset. The approach has been extended to convolutional filter pruning on the CIFAR-10 dataset using spatial global average pooling. The AEI approach has been compared to magnitude-based pruning methods like L1 and L2 norms and gradient-based pruning methods like SNIP and GraSP. The robustness of the proposed approach has been established by comparing the results over five random seeds. The AEI approach is proposed as a principled, interpretable, structure-aware pruning criterion rather than an accuracy-maximising method. Spectral analysis demonstrates that AEI is the only evaluated method that systematically targets structurally peripheral neurons, whereas magnitude-based methods prune broadly across the structural spectrum and GraSP actively removes structurally central neurons, an effect most pronounced on more complex datasets and deeper architectures (CIFAR-100, ResNet-20), where it causes substantial accuracy degradation at high sparsity. The AEI approach has been extended to the Hybrid approach by combining the AEI and L2 norms. The Hybrid approach has been seen to improve the accuracy gap at 40% sparsity from 5.46 to 2.26 percentage points over the state of the art. The robustness of the proposed approach has been established by conducting ablation studies on the robustness of the approach to the selection of the graph. The approach has been seen to be moderately robust to the selection of the graph with Spearman’s $\rho$ ≈ 0.70. The accuracy gap has been seen to be less than 0.2%. Full article

Automatic detection of stress in social media text holds promise for supporting digital mental health, but most existing Transformer-based approaches are opaque and computationally demanding. This work presents DR-Transformer, a Dual-Regularized Transformer that combines two complementary mechanisms: (i) a group sparsity penalty ($L_{2,1}/L_2$ elastic net) applied to the query and key projection matrices of every attention head, which encourages whole-row sparsity, producing more concentrated and inspectable attention patterns; (ii) a supervised contrastive loss on the [CLS] projection, which organizes the latent space according to the stress label. The architecture is intentionally lightweight (six layers, eight heads, 256-dim embeddings; ∼9.5 M parameters) and runs entirely on consumer-grade hardware (NVIDIA GTX 1660, 6 GB). Experiments on the publicly available Dreaddit dataset (binary stress classification, 2838 train/715 test segments) compare DR-Transformer against Logistic Regression, BiLSTM, a Standard Transformer of identical architecture, and MentalBERT. Across five seeded runs, DR-Transformer (Full) reaches $F_1=0.876$ (bootstrap 95% CI $0.852$–$0.898$), outperforming the Standard Transformer ($F_1=0.842$; McNemar $p<0.001$ with Bonferroni correction) and performing comparably to the much larger MentalBERT ($F_1=0.879$; $p=0.421$). Sparse regularization increases the fraction of near-zero attention weights (below $0.01$) from $0.215$ to $0.682$, while the supervised contrastive loss improves the silhouette score of [CLS] embeddings from $0.312$ to $0.483$. Dual regularization thus combines accuracy, efficiency, and structurally induced attention concentration in a single model which can be trained without specialized infrastructure. We use the term “interpretable” throughout in this restricted, structural sense—to refer to concentrated and inspectable attention—rather than in the sense of established causal or mechanistic faithfulness; this is only partially and indirectly supported by our token deletion analysis. Full article

Self-supervised graph collaborative filtering suffers from two geometric pathologies. In Dimensional Collapse, embeddings quietly collapse into a low-rank subspace, a well-documented but poorly solved problem. Semantic Collapse is more complex: stiff $L^2$-squared orthogonalization penalties that are supposed to push the embeddings apart end up ripping through the heavy-tailed community overlaps that contain the collaborative signal. Earlier studies attempted to mitigate sparsity by injecting static noise or structural perturbations, but such interventions did not pinpoint the root cause, i.e., the distortions in the global covariance geometry itself. In this paper, we propose a Huber-Contrastive Graph Convolutional Network (HCGCN) that combines a spatial message-passing backbone with an $O(1)$ contrastive augmentation overhead and a Relation-Aware Dual-View Gated Contrastive Network. The main novelty is a Huber Covariance Whitening module that imposes a geometry-aware threshold on the cross-correlation matrix of augmented views—below the threshold, the gradients follow an $L^2$ penalty (enforcing uniformity); above it, the penalty flattens to $L^1$ (protecting genuine semantic clusters from gradient explosion). This theoretically motivated dual-regime penalty actively preserves the macro-semantic topology of the graph while aggressively stamping out spurious noise correlations. The HCGCN is evaluated on the Yelp2018, Amazon-Book, and MovieLens datasets and performs significantly better than state-of-the-art baselines like LightGCN, SGL, SimGCL, and NESCL, especially under severe cold-start settings where covariance regulation proves most critical. Full article

Recommender systems face fundamental challenges, including extreme data sparsity and noisy rating observations. We propose L21RSVD, a robust latent factor model that employs L21 norm regularization within a Singular Value Decomposition framework to mitigate the impact of outliers while preserving low-rank structure. Unlike conventional L2-regularized approaches, our formulation induces group sparsity in the latent factor space, yielding more discriminative user and item representations. We derive three optimization variants: standard L21RSVD, L21RSVD without the squared term, and coefficient-free adaptive L21RSVD. Building upon these, we introduce a fusion strategy that adaptively aggregates predictions based on local data density. Extensive experiments on benchmark datasets demonstrate that L21RSVD substantially outperforms classical collaborative filtering and the SVD-type model. The proposed fusion model achieves state-of-the-art performance, reducing RMSE by up to 31.21% and MAE by up to 42.99% relative to baseline methods. Full article

Tunnel construction in tectonically active mountainous regions is frequently hampered by elevated geothermal conditions, threatening construction safety and long-term infrastructure performance. In the Yunnan–Guizhou Plateau, characterized by complex fault systems and intense hydrothermal activity, rigorous assessment of geothermal hazard susceptibility along tunnel corridors is of critical engineering importance. However, the sparsity of geothermal observational data renders conventional assessment approaches insufficient, as they fail to quantify predictive uncertainty, which is essential for reliable risk decision-making. To address this gap, this study proposes a three-stage framework that integrates multi-source data fusion with Monte Carlo-based uncertainty quantification, using the Nige Tunnel as a case study. Ten conditioning factors were incorporated, with temperature-weighted positive samples constructed from field-surveyed hot springs. Gaussian noise injection and fractal buffer randomization were applied across 500 Monte Carlo iterations of an L2-regularized logistic regression model, evaluated by leave-one-out cross-validation. The three-stage assessment framework achieved robust predictive performance (mean leave-one-out cross-validation area under the curve (LOO-AUC) = 0.824) under sparse-sample conditions. High and Very High susceptibility zones account for 14.2% of the study area, concentrated along fault traces and collocated with hydrothermal discharge locations, with the Nige Tunnel traversing predominantly High to Very High susceptibility zones. Fault distance emerges as the dominant predictive factor, surpassing heat flow and Moho depth, indicating that structural permeability is the rate-limiting control on geothermal fluid enrichment in fault-dominated systems. The findings offer scientific support and methodological insights for risk zoning and hazard mitigation design in tunnel engineering projects in comparable geological settings. Full article

Conversion from distributed acoustic sensing (DAS) measurements to geophone-equivalent data is important for integrating DAS into conventional seismic workflows. This is because most established seismic-processing algorithms are designed for particle-velocity or acceleration data, whereas DAS measures strain or strain rate. Recovering geophone-equivalent particle velocity from DAS strain-rate measurements requires inversion of a gauge-length-dependent spatial-difference operator, which can amplify measurement noise, particularly in field data with low signal-to-noise ratios (SNRs). Existing single-regularization methods often trade noise attenuation against waveform fidelity and the preservation of weak coherent events. To address these limitations, we propose an inverse reconstruction framework combining high-order L_p (HOLp) and overlapping group sparsity (OGS) regularizations. HOLp promotes a compact representation of second-order differences and suppresses incoherent fluctuations, whereas OGS exploits local coherence to reduce isolated artifacts and preserve weak continuous events. The resulting objective function is solved using the alternating direction method of multipliers, with iteratively reweighted L₁ minimization for the HOLp subproblem and a majorization–minimization strategy for the OGS subproblem. Numerical and field experiments confirm that the method restores amplitude and waveform fidelity under low SNR conditions, demonstrating robust and reliable DAS-to-geophone conversion. Full article

Hyperspectral unmixing (HU) aims to estimate endmembers and their corresponding abundances, a task commonly referred to as blind hyperspectral unmixing (BLU). Nonnegative matrix factorization (NMF) provides a unified framework for their joint estimation. It is widely assumed that more accurate endmember estimation leads to improved abundance estimation, enabling simultaneous optimization of both variables. However, this paper shows that, in practical noisy scenarios, the relationship between endmembers and abundances in NMF-based multi-variable joint optimization problems (NMF-based JOPs) is inherently coupled and significantly more complex, making it difficult to improve both estimation accuracies simultaneously. Furthermore, we demonstrate that the hard abundance sum-to-one constraint (ASC), commonly imposed in NMF-based JOPs, is inconsistent with realistic noisy conditions. To address these limitations, we propose a novel two-stage framework for BLU that decouples the refinement of endmembers and abundances. In the first stage, a strongly convex minimum-volume simplex model is employed to ensure robust and stable endmember extraction. In the second stage, we introduce a novel formulation, L₁_SoftASC, which promotes abundance sparsity and physical interpretability while improving convexity and robustness in abundance estimation. Experimental results on both synthetic and real benchmark datasets demonstrate that the proposed two-stage approach consistently outperforms existing single-stage NMF-based JOP methods in terms of both endmember and abundance estimation accuracy, while providing BLU with greater flexibility in handling ASC. Full article

The main aim of the airborne early warning (AEW) system is to search the potential targets in a large surveillance area. The underlying assumption is that the desired target signals only exist in a few range cells for space-time adaptive processing (STAP), i.e., targets (with certain look direction and Doppler) are sparsely distributed in the entire range cells and most of the range cells are target-free. By utilizing the sparsity of the target distribution, we propose a new STAP method by minimizing the l1-norm of the output magnitude. Unlike conventional STAP methods, which exclude the cell under test from the training samples to avoid target self-nulling, our method processes the cell under test (CUT) and the training samples simultaneously without sample selection. Moreover, to achieve robustness against target steering vector mismatch, we constrain the l1-modulus of the response of any steering vector within a rhombus uncertainty set to exceed unity. Additionally, based on a new definition of the l1-norm of a complex-valued vector, the original nonlinear programming problem can be transformed into a linear programming problem. On the other hand, unlike the slide window processor (SWP) whose weights need to be updated for each range cell, the adaptive weight of our method for a block of samples requires no updating. Consequently, the computational complexity of the proposed method is much lower than that of conventional STAP methods. Finally, since the CUT is used to compute the STAP weights, our method can also suppress the discrete interference. The robustness, computational effectiveness and superiority of the proposed STAP method are verified based on simulated data and the MCARM data. Full article

Grant-free non-orthogonal multiple access (NOMA) enables communication without a scheduling process. Base station (BS) must detect active users without knowing their number, a challenge that also occurs in grant-free NOMA–Internet of Things (IoT). Device detection in grant-free NOMA-IoT can be considered as signal reconstruction in compressive sensing (CS). To address this limitation, we propose extended sparsity estimation- orthogonal matching pursuit (ESE-OMP) to detect active devices in single measurement vector (SMV) and multiple measurement vector (MMV) problems for grant-free NOMA-IoT systems, a reconstruction method in CS that operates without prior knowledge of the sparsity level, which corresponds to the number of active devices. The algorithm iteratively detects active devices by monitoring the absolute difference in $l_1$-norm of successive residuals, terminating when the change falls below a predefined threshold $\epsilon$. ESE-OMP is evaluated under various grant-free NOMA-IoT systems, irregular low-density spreading-orthogonal frequency division multiplexing (LDS-OFDM), regular LDS-OFDM, and pattern division multiple access (PDMA) systems. When the signal-to-noise ratio (SNR) is 10 dB for the SMV problem with static active device composition, the regular LDS-OFDM system achieves a bit error rate (BER) of $2.95\times {10}^{-4}$, while irregular LDS-OFDM and PDMA systems achieve BERs of $3.78\times {10}^{-3}$ and $1.79\times {10}^{-2}$, respectively. The smaller the number of active devices, the better the performance of ESE-OMP. Full article

Intrusion Detection Systems (IDSs) remain essential for securing modern network infrastructures, where traffic data are often high-dimensional and contain redundant or weakly informative attributes. This study proposes a hybrid feature selection approach that combines Information Gain with L1-regularized selection to construct a compact and informative representation of the UNSW-NB15 dataset. The method applies relevance-based filtering followed by sparsity-driven refinement within a leakage-aware pipeline, in which preprocessing and feature selection are derived exclusively from the training data. Under a reduced six-class configuration, the proposed approach reduces 42 candidate predictors to 21 traffic-related features. Across multiple classifiers, Random Forest + IGL1 achieved the best performance, with an accuracy of 0.8432 and an F1-score of 0.8376, while MLP and Gradient Boosting also remained competitive. These findings indicate that the selected features preserve consistent discriminative patterns rather than favoring a single classifier. Overall, the study highlights the importance of leakage-aware evaluation for producing reliable, reproducible intrusion detection results. Future work will extend the analysis to the full multi-class setting and examine applicability in real-time or streaming environments. Full article