Multimodal automatic modulation recognition (MAMR) has emerged as a prominent research area. The effective fusion of features from different modalities is crucial for MAMR tasks. An effective multimodal fusion mechanism should maximize the extraction and integration of complementary information. Recently, fusion methods based on cross-modal attention have shown high performance. However, they overlook the differences in information intensity between different modalities, suffering from quadratic complexity. To this end, we propose an efficient Alignment before Fusion Transformer Network (AbFTNet) based on an in-phase quadrature (I/Q) and Fractional Fourier Transform (FRFT). Specifically, we first align and correlate the feature representations of different single modalities to achieve mutual information maximization. The single modality feature representations are obtained using the self-attention mechanism of the Transformer. Then, we design an efficient cross-modal aggregation promoting (CAP) module. By designing the aggregation center, we integrate two modalities to achieve the adaptive complementary learning of modal features. This operation bridges the gap in information intensity between different modalities, enabling fair interaction. To verify the effectiveness of the proposed methods, we conduct experiments on the RML2016.10a dataset. The experimental results show that multimodal fusion features significantly outperform single-modal features in classification accuracy across different signal-to-noise ratios (SNRs). Compared to other methods, AbFTNet achieves an average accuracy of 64.59%, with a 1.36% improvement over the TLDNN method, reaching the state of the art.

Introduction: In the big data era, healthcare data has grown exponentially, presenting opportunities to explore the pathogenesis of heart disease. Clarifying the correlations between health indicators and heart disease is crucial for early prevention. This study employs ensemble learning to identify the key influencing factors, assisting clinicians in understanding the pathogenesis and enhancing prediction strategies. Methods: A two-layer stacking ensemble model is proposed, integrating Naive Bayes, Decision Trees, CatBoost and Gradient Boosting Trees to enhance prediction accuracy. To address ensemble models’ complexity and poor interpretability, the SHAP technique is introduced to visualize the decision-making logic of the ensemble model. Results: Experimental results show that the stacking model achieved 86.69% accuracy, 87.14% weighted precision, 86.69% weighted recall, and 86.91% weighted F1-score. It balances precision and recall, unlike single learners that prioritize one over the other. Global interpretive analysis demonstrates that age, sleep duration, self-rated health status and BMI are critical factors in assessing cardiovascular risk. Local interpretive analysis is conducted to evaluate the contribution of each feature to the prediction results of individual samples. Discussion: The stacking model’s superior performance demonstrates that ensemble learning can overcome the limitations of single learners. Additionally, key predictive factors are identified: maintaining an average sleep duration of 7-8 hours significantly reduces heart disease risk, while advanced age and poor health status increase susceptibility. This study provides a reliable predictive tool for personalized heart disease prevention and treatment.

Accurate truncation of unbounded domains is a central challenge in time-domain electromagnetic (EM) systems. While perfectly matched layers (PMLs) can deliver excellent absorption, they typically require multiple grid layers and auxiliary variables, increasing runtime and memory costs. This work introduces a single-layer neural complete radiation boundary condition (NP-SSM-CRBC) that replaces a thick absorbing boundary with a one-cell boundary model. Instead of explicitly evolving auxiliary fields as in conventional PML or convolutional PMLs (CPMLs) formulations, the proposed approach employs a neural state-space model to predict the boundary ghost-cell responses directly. The method learns an effective, time-causal boundary operator and is designed to remain stable over long time horizons through explicit physical constraints. Experimental results demonstrate that the proposed method achieves low reflection levels with substantially reduced computational overhead. These results provide a foundation for extension to larger three-dimensional problems and more complex boundary geometries.

1. Cyclin-dependent kinases (CDKs) constitute a fundamental family of serine/threonine kinases whose catalytic activation is strictly contingent upon binding to specific regulatory subunits termed ...