Search Results (258)

With the widespread adoption of intelligent terminals, edge devices, and distributed information systems in the financial domain, financial security sensing data exhibit multisource heterogeneity, dynamic temporal patterns, and high privacy sensitivity. Traditional centralized anomaly detection methods are no longer able to simultaneously satisfy the requirements of cross-institutional or cross-node collaborative modeling, client data privacy protection, and robust monitoring of transaction and system anomalies. To address this challenge, a data-local federated deep anomaly detection framework has been proposed for distributed financial security sensing systems. Initially, a local deep financial security sensing representation module is constructed to perform temporal encoding and attention-based modeling on multisource financial signals, including terminal operation status, network transaction communication, backend server operation, identity authentication, and anomaly alerts, thereby extracting representations relevant to anomalous behaviors. Subsequently, a data-local federated optimization and personalized aggregation mechanism is developed to enable cross-node knowledge sharing without transmitting raw transaction or client data, while local personalized detection heads are employed to adapt to non-independent and identically distributed (non-IID) financial institution data. Furthermore, an adversarially robust security detection and trust-aware aggregation strategy is introduced to enhance model stability under input noise, feature masking, anomaly camouflage, and potential malicious client updates. Experimental results demonstrate that the proposed method achieves an Accuracy of $92.37\%$, a Precision of $89.41\%$, a Recall of $88.26\%$, an F1-score of $88.83\%$, an AUC of $93.06\%$, and a PR-AUC of $89.15\%$ in the primary financial anomaly detection task, significantly outperforming baseline methods such as Isolation Forest, Autoencoder, LSTM, Transformer, FedAvg, FedProx, SCAFFOLD, and MOON. In robustness experiments, the method attains F1-scores of $87.95\%$, $86.42\%$, $86.88\%$, $84.57\%$, $86.73\%$, and $83.91\%$ under Gaussian noise, feature masking, temporal shift, adversarial perturbation, and $20\%$ and $30\%$ malicious client scenarios, respectively. Ablation studies further confirm the effectiveness of local representation learning, personalized federated optimization, adversarial training, and trust-aware aggregation mechanisms. Overall, the proposed approach provides an efficient intelligent anomaly detection solution for financial AI security monitoring scenarios characterized by data localization requirements, node heterogeneity, and attack perturbations. Full article

Crack opening and reinforcement stress are two complementary indicators of the service state of reinforced concrete hydraulic structures, yet they are often predicted separately. This study develops a data-driven multi-task temporal fusion framework for joint 48 h ahead prediction of dam crack responses and rebar stress using multi-source monitoring data. The measured data comprise five crack-monitoring series, five rebar stress series, local temperature channels, reservoir water level, antecedent rainfall, and an auxiliary environmental signal over approximately four years. Target responses are aligned only at common measured timestamps; no synthetic target observations are introduced. A simplified engineering layout and plan-based crack–rebar distances are further used to examine whether an explicit spatial prior can strengthen the shared temporal representation without introducing synthetic target values. A residual multi-task temporal fusion network (MTTF-Net) is proposed with a shared Transformer encoder, attention pooling, task-specific decoders, and a response-continuity regularization term. The model is compared with persistence, Ridge regression, random forest, Extra Trees, XGBoost, and GRU baselines under a chronological train/validation/test split. For the independent test period, Ridge regression obtains the lowest overall RMSE (2.2968), whereas MTTF-Net provides the lowest crack RMSE (0.0141), the lowest overall MAE (1.0035), and the second-best overall RMSE (2.3813). Distance-informed ablation, denoted as MTTF-Net-S, remains close to MTTF-Net in macro-averaged $R^2$ but is not superior in the overall test metrics, indicating that the available horizontal distances are valuable engineering metadata but cannot replace richer three-dimensional structural connectivity. These results indicate that the monitoring data contain a strong linear autoregressive component, while multi-task temporal fusion improves nonlinear crack response prediction and remains competitive for stress forecasting. The source code is prepared as a public implementation package, whereas the measured monitoring dataset is subject to data owner restrictions. Full article

Intelligent sensors and Internet of Medical Things (IoMT) platforms are rapidly changing smart healthcare by enabling continuous capture of physiological, behavioral, and clinical events outside conventional hospital settings. Yet the value of connected sensing depends on more than signal acquisition alone. A practical remote-monitoring ecosystem must also convert sensor alerts, clinician-facing summaries, and historical electronic clinical records (ECRs) into ranked evidence that supports care decisions. This study reframes a large-AI clinical retrieval model as the intelligence layer of an edge–cloud IoMT architecture. The proposed framework combines Transformer-Based Sequence (TBS) encoding, BioBERT-driven representation learning, explicit retrieval, and domain-guided re-ranking to connect sensor-originated narratives, patient records, and clinician queries. The empirical evaluation is conducted on Medical Information Mart for Intensive Care III (MIMIC-III) and i2b2, two de-identified clinical text benchmarks that approximate the documentation layer of real-world remote patient monitoring. Compared with strong baselines, including DeepBio, UniT2T, Web4IR, A2A-API, CoLTiD, VLRG, ColBERT, DeepSDH, BiRex, and DL4BTM, the proposed model achieves the best overall performance, reaching F1/Pre/NDCG scores of 0.8399/0.8338/0.5235 on MIMIC-III and 0.8090/0.8100/0.5129 on i2b2. Ablation experiments confirm the importance of exploratory data adaptation, critical feature modeling, critical token learning, cross-disciplinary supervision, and data-driven regularization. Parameter sensitivity analysis shows stable behavior for beta values greater than or equal to 1, with the strongest results at beta = 5. The study concludes that large-AI retrieval can strengthen the clinical interpretation layer required for IoMT-enabled remote monitoring, while future work should validate the approach on live multimodal sensor streams and privacy-preserving deployments. Full article

Automatic Modulation Classification (AMC) is essential for waveform-level signal characterization. It supports spectrum sensing, signal identification, and adaptive resource allocation in cognitive radio and next-generation wireless systems. However, channel impairments such as multipath propagation, frequency offset, fast fading, and noise degrade modulation signatures, making reliable AMC challenging. Existing deep learning-based approaches often rely on purely data-driven learning, leading to insufficient modeling of modulation-relevant features, loss of transient characteristics, and limited exploitation of hierarchical relationships among modulation types. To address these issues, this paper proposes PHM-Net, a physics-informed hierarchical multi-scale network for robust AMC. The model employs a hierarchical backbone with residual encoder blocks. A Transient Feature Gating (TFG) module enhances modulation-relevant representations, a Cross-Resolution Signal Aggregation (CRSA) module fuses multi-stage features, and a Physics-Informed Hierarchical Loss (PI-HL) enforces consistency between coarse- and fine-grained predictions. Experimental results on three benchmark datasets (RML2016.10a, RML2016.10b, and RML2018.01a) show that PHM-Net consistently achieves the highest average accuracy among all compared models. On RML2018.01a, which contains 1024-sample sequences and 24 classes, PHM-Net achieves an average accuracy of 64.59% and a best-case accuracy of 98.42%, surpassing AMC_Net by 11.14 and 17.09 percentage points and CNN-Transformer by 9.43 and 11.15 percentage points, respectively. PHM-Net provides a robust and interpretable solution for AMC under complex channel conditions. Full article

The long-term stability of submerged sandbars and protected shorelines in large alluvial rivers depends on the serviceability of flexible mattresses installed on the riverbed. Distributed fiber optic sensing is one of the few practical methods for monitoring deformation along these underwater systems over engineering-scale distances. Yet BOTDR-derived strain-difference profiles are often heavily contaminated by noise and rarely have reliable clean references. To address this issue, this study develops TrendBlend-BSFormer, a blind self-supervised denoising framework for in situ BOTDR strain data from underwater flexible mattresses. The framework combines four key features: blind-spot masking, a one-dimensional encoder decoder backbone, a Transformer bottleneck for long-range spatial dependence, and a multi-scale trend-detail blending branch with dual signal-noise heads. The framework was validated using annual and daily BOTDR field data from the Yudaizhou shoreline protection project in the Yangtze River, containing 9343 and 9875 valid measurement points, respectively. TrendBlend-BSFormer achieved pseudo-SNR/RMSE/MAE values of 14.22 dB, 15.03 με and 12.05 με for the annual data set and 5.32 dB, 8.02 με and 6.45 με for the daily data set, improving the pseudo-SNR by 1.45 dB and 2.95 dB relative to the published BiLSTM-CNN benchmark. It also reduced the high-frequency energy ratio from 0.172 to 0.011 for the annual data and from 0.424 to 0.112 for the daily data. The denoised profiles suppress isolated spikes while preserving mechanically plausible peaks, valleys, and short-range fluctuations, indicating that blind self-supervised denoising can provide a more physically credible strategy for BOTDR-based monitoring in complex underwater environments. Full article

Accurate volatility forecasting and timely risk early warning are foundational requirements of financial risk management: Value-at-Risk estimates, portfolio risk limits, derivative hedging ratios, and stress-test scenario calibrations all depend on forward-looking volatility signals that remain reliable when market conditions depart from average. This paper develops a public-data causal multiscale wavelet spillover learning (CMWSL) framework that jointly addresses stock-index volatility forecasting and high-volatility early warning under strict walk-forward evaluation. CMWSL integrates three components: a heterogeneous autoregressive (HAR) persistence block as the dominant linear baseline, causal stationary wavelet transform (SWT) summaries that encode within-index multiscale market dynamics, and a cross-index spillover layer that tests whether medium- and long-scale wavelet energy from peer indices carries incremental risk-relevant information. The empirical analysis covers the S&P 500, Nasdaq-100, and Dow Jones Industrial Average over a 2513-step out-of-sample evaluation period from 2016 to 2025, with forecast horizons $h\in \{1,5,10\}$ and OHLC-based volatility targets. All preprocessing, wavelet decomposition, calibration rules, and warning thresholds are re-estimated inside each rolling training window to eliminate look-ahead bias. HAR remains the strongest average model in the main Rogers–Satchell specification, confirming that daily index volatility risk is highly persistence-driven. The multiscale extension delivers statistically significant improvements at longer horizons, in richer public macro-financial information environments, and under the Parkinson target. Clark–West tests detect significant spillover gains in five of nine index–horizon cells (CW $=4.83$, $p<0.001$ for S&P 500 at $h=10$). Critically, tail-conditioned and rolling-window diagnostics show that multiscale and cross-index gains concentrate in upper-volatility regimes and synchronized stress episodes—precisely the conditions in which risk management decisions are most consequential. For market-risk early warning, a logistic classifier built on the same causal feature pipeline delivers the most stable precision–recall performance across all settings, providing an interpretable and operationally auditable alert mechanism suitable for practical risk monitoring. Full article

Early stopping is a standard form of implicit regularization in neural sequence models, but criteria based solely on validation loss can become unstable or weakly informative in noisy, non-stationary, or weakly separated regimes. We propose Symbolic Early Stopping (SES), a representation-aware hybrid stopping criterion that monitors the evolution of validation hidden-state organization during training. At each epoch, SES constructs a Mapper-based symbolic abstraction of hidden representations extracted from a fixed monitored layer, transforms latent trajectories into symbol sequences, and summarizes them through a compact set of symbolic–dynamic descriptors capturing sequential complexity, transition uncertainty, and geometric dispersion. These descriptors are aggregated into a single symbolic stability score, which is combined with validation-loss monitoring to detect convergence of the learned representation. We evaluate SES on recurrent, bidirectional recurrent, and encoder-only Transformer architectures across multiple time-series regimes with different levels of structural regularity and noise. The results indicate that SES frequently terminates training substantially earlier than conservative loss-based baselines while preserving a competitive quality–efficiency trade-off relative to oracle validation-based stopping. Robustness experiments under additive input noise show that the symbolic monitoring signal remains informative under moderate perturbations, although its advantage is not uniform across all datasets and model classes. A layer-wise analysis further suggests that useful stopping signals may emerge before the final validation curve fully stabilizes, reflecting earlier organization of latent representations. Overall, SES provides an interpretable and computationally tractable framework for representation-level early stopping in neural sequence modeling. Full article

Chronic stress and depression-adjacent emotional states affect over 970 million people worldwide, yet their continuous, objective detection from physiological signals remains unsolved, particularly when cues are subtle and conventional approaches fail. Single modality classifiers capture only part of the picture, and binary valence/arousal formulations collapse emotionally distinct states into the same category, leaving conditions like sadness and depression characterised by Low-Valence Low-Arousal (LVLA) responses without reliable detection. A hybrid deep learning model combining Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Transformer encoders was developed to jointly classify four emotional quadrants from multimodal physiological data in the Database for Emotion Analysis using Physiological signals (DEAP) dataset. Electroencephalography (EEG), Galvanic Skin Response (GSR), Blood Volume Pulse (BVP), and respiration signals from 32 participants were preprocessed using fourth-order Butterworth filtering, trained with Adam optimisation, and evaluated through an 80/20 stratified split with five-fold cross-validation. The system achieved 91.2% four-quadrant valence–arousal accuracy (95% CI: 89.1–93.3%), with LVLA recall reaching 91.3%, outperforming all partial-hybrid variants. These findings demonstrate that hierarchical, attention-based fusion of physiological modalities can reliably distinguish stress from depression-adjacent states, offering a practical pathway toward continuous, non-invasive mental health monitoring on wearable platforms. Full article

The era of zero-sum competition calls for e-commerce platforms to shift focus toward micro-market resilience. Existing research has split into two traditions: diagnostic studies offer detailed analyses of market failure but lack systemic application, while engineering studies develop deployable tools yet suffer from opaque mechanisms and hidden risks. This paper proposes the Signal–Belief–Decision (SBD) framework to bridge this divide, with the Signal layer transforming private information into verifiable public knowledge, the Belief layer aggregating dispersed signals into shared consensus, and the Decision layer encoding enforceable rules for incentive compatibility. Using an extended signaling game, we diagnose six vulnerability dimensions (VD1–VD6) that destabilize markets. Agent-based modeling then allows us to distill four design principles (DP1–DP4) that inform governance configuration. The SBD framework provides a middle-range theoretical architecture that reorients platform governance from reactive tooling to proactive, consumer-centric design. Full article

Traffic flow prediction is important for route planning, signal control, and traffic guidance. However, traffic-state sequences usually exhibit non-stationarity, periodicity, and complex temporal dependencies, which makes it difficult for traditional statistical methods and single deep learning models to simultaneously capture short-term local fluctuations and long-term evolutionary trends. To address this issue, this paper proposes a dual-stream latefusion CNN-LSTM with an adaptive gated shortcut for traffic flow prediction, denoted as AGS-CNN-LSTM. The proposed method does not aim at explicit spatial-topology modeling; instead, it focuses on improving the fusion mechanism of CNN-LSTM-based models under settings without graph-structure constraints. Based on two public datasets, PeMS-BAY and PeMSD8, this study constructs multi-step prediction tasks with horizons of 15 min, 30 min, 60 min, 90 min, and 120 min and compares the proposed model with MLP, SimpleRNN, 1DCNN, LSTM, Serial CNN-LSTM, CNN-LSTM-Attention, BiLSTM-Attention, TCN-LSTM, Transformer Encoder, DLinear, and DS-CNN-LSTM (w/o Gate). The experimental results show that AGS-CNN-LSTM does not consistently achieve the best performance across all datasets, prediction horizons, and evaluation metrics. Nevertheless, it performs close to the best baseline models on the 30 min and 60 min tasks of PeMS-BAY and achieves competitive RMSE and R² results on the 15 min, 30 min, and 60 min tasks of PeMSD8. Further ablation experiments indicate that the adaptive gated shortcut can enhance the predictive capability of the dual-stream late-fusion structure in some scenarios, although its benefits are dependent on the dataset and prediction horizon. Overall, the proposed model is more appropriately regarded as a lightweight fusion-mechanism improvement for CNN-LSTM-based models under settings without explicit graph-structure constraints, rather than a comprehensive replacement for complex graph neural networks, Transformerbased models, or models incorporating multiple external factors. Therefore, the findings should be interpreted as proof-of-concept evidence for a lightweight CNN-LSTM fusion enhancement under constrained non-graph-input settings, rather than as evidence of broad generalizability in complete road-network-level traffic forecasting. Full article

Cone-beam computed tomography (CBCT) is widely used in image-guided radiotherapy because of its low radiation dose and on-board acquisition capability. However, CBCT images often suffer from scatter artifacts, increased noise, reduced soft-tissue contrast, and inaccurate Hounsfield Unit (HU) values, which limit their direct use for accurate dose calculation and quantitative analysis. To address this limitation, we propose a CBCT-to-CT synthesis framework based on 2.5D context encoding (concatenating five adjacent slices along the channel dimension) and latent-space variational diffusion. The proposed method combines a Vector Quantized Variational Autoencoder (VQ-VAE) and a U-shaped Vision Transformer (U-ViT)-based latent-space Variational Diffusion Model (VDM) to translate CBCT images into synthetic CT (sCT) images in a compressed latent space. To incorporate inter-slice anatomical context while preserving the computational efficiency of 2D processing, five adjacent CBCT slices are concatenated along the channel dimension and used as input. We evaluated the proposed method on the SynthRAD2025 paired CBCT-CT dataset covering head-and-neck, thoracic, and abdominal regions. Under the provided benchmark setting, quantitative evaluation on the validation set showed that the proposed 2.5D model improved peak signal-to-noise ratio (PSNR) from 25.39 dB to 27.44 dB (averaged across regions), structural similarity index measure (SSIM) from 0.813 to 0.846, reduced mean squared error (MSE) from 0.00313 to 0.00200, and lowered Fréchet inception distance (FID) from 1009.33 to 869.53 compared with the 2D baseline. Qualitative results also showed improved anatomical consistency and reduced artifact-related distortions. These findings suggest that neighboring-slice context can enhance HU fidelity and overall image quality in a computationally practical synthesis framework, supporting the usefulness of efficient AI-based cross-modality reconstruction for radiotherapy-related imaging workflows. Full article

Electroencephalography (EEG) has emerged as a promising non-invasive tool for the diagnosis of neurodegenerative disorders, and artificial intelligence (AI) has shown significant potential in this domain, as demonstrated by recent studies. However, strong inter-subject variability remains a major challenge, limiting the ability of AI-based models to learn disease-specific features that generalize across individuals, thereby hindering the development of clinically deployable subject-independent systems. In this work, we propose a cross-subject, AI-based EEG classification framework to distinguish between Alzheimer’s disease (AD), Creutzfeldt–Jakob disease (CJD), and healthy control subjects using clinical EEG data collected from a local hospital. A lightweight hybrid deep learning model is developed, combining a two-layer one-dimensional convolutional neural network with a two-layer Transformer encoder to capture both local temporal patterns and long-range dependencies in EEG signals. The proposed model achieves an average classification accuracy of $97\%$, representing a $3\%$ improvement over a baseline model evaluated on a cohort of 36 subjects. To assess deployment feasibility in real-time clinical settings, the trained model is implemented and evaluated on an edge-AI platform (NVIDIA Jetson AGX Orin), demonstrating energy efficiency for the inference with a compact model footprint. These results indicate that the proposed approach provides an accurate, efficient, and practically deployable solution for subject-independent EEG-based classification of neurological disorders. Full article

In recent years, channel state information (CSI)-based sensing technology has gradually attracted widespread attention as a contactless and low-cost approach for robotic arm motion understanding. Despite continuous progress in CSI-based human sensing, existing methods of robotic motion sensing still face two key challenges when directly applied to robotic motion sensing: (1) CSI perturbations induced by robotic arm motion are weak and locally distributed, making fine-grained feature extraction difficult. (2) Discriminative information in long robotic arm motion sequences is sparsely concentrated in a few key intervals, and its adaptive temporal selection and enhancement remain challenging. To address the above challenges, this paper proposes an efficient multi-stage robotic arm motion recognition method (named MSPoolNet). The proposed method consists of three key modules: an adaptive temporal downsampling module, a temporal gating module, and a Transformer-based feature encoding module. Specifically, the adaptive temporal downsampling module processes the raw CSI signal at the input stage to achieve local pattern extraction. The temporal gating module adaptively reweights temporal features, dynamically highlighting key temporal segments while suppressing irrelevant information. The proposed Transformer-based feature encoding module replaces conventional self-attention with pooling operations, enabling global information interaction and fine-grained feature representation in a computationally efficient manner. Extensive experimental results demonstrate that the proposed method achieves state-of-the-art performance on two representative public datasets, maintaining a compact model size with an accuracy exceeding 99%. Full article

Automatic modulation classification (AMC) is a key component of spectrum awareness, cognitive radio, and signal intelligence, enabling receivers to identify modulation schemes from noisy in-phase and quadrature (IQ) observations. Traditional approaches rely on likelihood-based methods or handcrafted feature extraction, which often struggle under channel impairments and real-world variability. Recent advances in deep learning enable models to learn directly from multiple signal representations, including raw IQ samples, engineered features, and time–frequency or constellation-based encodings, improving adaptability across diverse signal conditions. This paper presents a structured review of deep learning approaches for AMC, including CNNs, RNN/LSTM models, and transformer-based architectures, with a focus on performance, robustness, and system-level trade-offs. We analyze how representation choices, dataset design, and evaluation protocols influence reported results, and highlight key challenges such as domain shift, low-SNR environments, and multi-signal interference. Finally, we outline future directions focused on improving generalization, integrating classical signal processing with learning-based methods, and enabling efficient deployment in real-world and resource-constrained systems. Full article

Depression is a critical global public health challenge, and the demand for accurate automated detection methods has generated considerable research interest in Transformer-based models. Despite their substantial promise, a comprehensive investigation into their architectural efficacy, intrinsic mechanisms, and barriers to practical implementation remains lacking. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, this systematic review was conducted across six databases (IEEE Xplore, Elsevier, Springer, MDPI, PubMed, and arXiv). The final search was performed in October 2025, covering English-language empirical studies published between 2020 and 2025 that employed Transformer-based architectures for depression detection. Risk of bias and methodological quality were independently appraised by two authors using a six-dimension structured rubric, with disagreements resolved by a third author. Findings were narratively synthesized given substantial cross-study heterogeneity. This systematic review analyzed 46 studies and provided the first comprehensive, mechanism-level, architecturally stratified comparison of encoder-only, decoder-only, hybrid, and multimodal fusion paradigms, examining self-attention dynamics and transfer learning strategies. Since 2019, these frameworks have evolved from text-centric approaches to advanced multimodal systems. Encoder-only models show consistently strong results in high-throughput text-based screening, decoder-only models demonstrate stronger few-shot learning capabilities, hybrid architectures show the highest observed median performance in clinical interview settings across the reviewed studies, and multimodal fusion systems offer complementary advantages when heterogeneous signal integration is critical. These trends are task-contextualized and should not be interpreted as unconditional rankings, given heterogeneity in evaluation metrics and tasks across studies. Nonetheless, four principal challenges hinder clinical translation: overreliance on self-reported data, cross-linguistic bias, absence of uncertainty quantification, and substantial computational overhead. Future efforts should shift from incremental benchmark improvements toward clinical utility through standardized psychiatric validation, uncertainty-aware architectures, fairness-enforced training across diverse populations, and the integration of Transformer-based models with wearable and mobile health data to improve detection stability and reduce translational risk. This systematic review was registered on the Open Science Framework (OSF; DOI: 10.17605/OSF.IO/SYF9N). This research was funded by the Faculty of Information Science and Technology and by Universiti Kebangsaan Malaysia under Grant TAP-K014364. Full article