Federated learning enables global model training in a distributed manner without requiring clients to upload private data. However, the presence of malicious clients uploading incorrect model updates can severely degrade the performance of the global model. Existing methods often rely on limiting the number of malicious clients or require access to additional clean datasets. To address these limitations, we propose a secure aggregation algorithm named FedMCP to detect and remove malicious clients. FedMCP first constructs judgment vectors based on parameter importance. It then distinguishes benign from malicious clients based on the similarity between their judgment vectors. Moreover, the discrimination between benign and malicious clients is achieved by incorporating the historical distribution characteristics of the judgment vectors from known benign clients. Finally, the Isolation Forest algorithm is employed to remove malicious models that mimic the judgment vectors. Even in individual rounds where a large number of malicious clients participate in training, FedMCP can still accurately distinguish between benign and malicious clients. FedMCP maintains high identification accuracy for both benign and malicious clients, even under heavy adversarial participation. Extensive experiments across multiple datasets and models confirm that FedMCP effectively identifies and excludes malicious clients, achieving superior robustness and performance.

Existing remote sensing image registration methods face an inherent trade-off among accuracy, efficiency, and generalization capability. To address this issue, this paper proposes a novel hierarchical Mamba network, named GeoMamba. This network introduces three core innovations: a Pyramid Mamba Block (PMB) to effectively capture hierarchical spatial features, an innovative Feature Dual-Fusion Module (DFM) to achieve efficient feature interaction with linear complexity, and a Dual Regression Constraint (DRC) strategy to enforce geometric consistency. Comprehensive experiments on the large-scale Aerial Image dataset show that GeoMamba’s performance comprehensively surpasses current state-of-the-art (SOTA) methods, reducing the Root Mean Square Error (RMSE) by up to 28.1% compared to the strongest baseline, while maintaining an average registration time of 0.3 seconds. More importantly, when directly applied to the SUIRD dataset for zero-shot generalization testing, the model demonstrates exceptional robustness and adaptability, proving that it has achieved a new SOTA level in terms of accuracy, efficiency, and generalization.

Severe optical signal deterioration in dim environments significantly bottlenecks existing image enhancement algorithms. To tackle complex, sensor-induced corruptions without relying on paired reference data, we propose a novel unsupervised framework: the Distribution Gap Feedback Network (DGF-Net). Our method conceptualizes real-world sensor degradation as an optimizable domain divergence. Through a cyclic estimation-and-injection loop, DGF-Net dynamically isolates specific degradation patterns from authentic low-light captures and feeds these residues back to calibrate the synthetic training stream. Additionally, a contextual random masking prior is integrated to fortify spatial reconstruction and prevent overfitting to localized noise. Extensive evaluations confirm DGF-Net establishes new state-of-the-art benchmarks. Quantitatively, it yields a 1.58 dB PSNR improvement over the runner-up technique on paired datasets and eclipses the unsupervised Zero-DCE method by over 9 dB. The architecture demonstrates exceptional robustness in resurrecting structural fidelity and chromatic balance under extreme real-world dark conditions, reliably benefiting downstream computational vision systems.

In recent years, multivariate time series anomaly detection has become an important research topic in the field of anomaly detection. In Industrial Internet of Things (IIoT) systems, the collected data may be affected by internal failures, external disturbances, or other adverse factors. In such cases, appropriate anomaly detection methods are required to ensure the stable operation of the system. However, existing methods based on reconstruction, prediction, or hybrid approaches often suffer performance degradation when anomalies are present in large amounts of training data, as these anomalies can negatively impact the training process. To address this challenge, we propose a dynamic masking strategy in both temporal and spatial dimensions. We develop a time series imputation framework based on a diffusion model that integrates Graph Neural Network (GNN) and Transformer architectures. This framework, termed Spatial-Temporal Dynamic Mask Diffusion for Anomaly Detection (STDMD-AD), incorporates a dynamic masking mechanism: temporally, reconstruction errors are used to mask data by randomly concealing values with higher errors; spatially, attention is applied to mask channels that are more likely to contain anomalies during training. Experiments on five real-world datasets demonstrate that the proposed method outperforms existing benchmarks and achieves state-of-the-art anomaly detection performance.

Health emergencies need to face the potential challenges about the public psychological healthcare and heart disease concurrent occurrence probability, especially for the large-scale pandemic disease spread situation. The complex bidirectional relationship between cardiovascular and psychological diseases highlights the importance of appropriate combination drug strategies, because of psychological therapy medicines could affect the heart sympathetic and parasympathetic nerve responses, at the same time, there may also cause a counter-effect. However, existing drug interaction prediction models focus only on the two-by-two drug relationship, which is difficult to meet the actual clinical needs for synergistic combination drugs. To address this, this paper proposes a Loewe score dual-guided diffusion model for synergistic drug combination generation aimed at generating drug combinations with high synergistic potential. It introduces drug text description information and Loewe synergistic score information into the training process of diffusion models, uses Loewe scores for conditional bias and weighted loss, and doubleguided denoising network learning to generate drug combinations with high synergy potential. The experimental results show that our proposed model is able to generate reasonable and reliable drug combinations, which provides a new perspective and an effective tool for solving the problem of clinical combination drug selection.

Identifying Systemic Lupus Erythematosus (SLE) and Antiphospholipid Syndrome (APS) has been challenging, because of the complex symptom overlap and biological differences between these two autoimmune diseases. To address this issue, this study proposes a text-based gene unsupervised clustering model and a gene dense subgraph mining algorithm to delve into their differences and similarities at the genetic level. The proposed gene unsupervised text clustering model is used for discovering the gene cross characteristics and significant differences between SLE and APS related gene groups, which is calculated by extracting semantic features from medical textual datasets. Then, the proposed gene dense subgraph mining algorithm is well identified the key and tightly connected gene sets that may play a critical role in disease mechanisms over largescale complex network, which would help to deep understand the correlations between them. The experiment results show that 10 typical functional gene groups (SLE 6 and APS 4 clusters) are clearly classified, and gave a detailed comparison analysis with the baseline methods and gave a semantic explanation for mined key related genes.

Type 2 diabetes mellitus (T2DM) requires multitargeted treatment, with drug efficacy tied to structure and function, though “structure-function divergence” exists. Drug similarity analysis and drug-target interaction (DTI) network construction are key for exploring functional associations and predicting targets. Thus, this study proposes an integrated framework for T2DM drugs that combines structural features and functional annotations, integrating similarity analysis and DTI network construction to explore links between structural features and binding affinity, aiding drug repurposing and target prediction. Via DTI networks, we calculated weighted similarity for small molecules using molecular fingerprints and functional annotations, applied global alignment with chain length weighting for double-stranded proteins, and used optimized local alignment for fusion proteins. The results indicate that the similarity between small molecule drugs of the same class is higher than that between drugs of different classes. The conservatism of core functional domains in protein drugs determines their selectivity for targets.

This study proposes a drug recommendation model based on a multi-objective optimization algorithm, aiming to provide personalized medication strategies for Systemic Lupus Erythematosus and Antiphospholipid Syndrome patients, taking into account individual differences and medication risks. Using a multi-objective optimization algorithm, the model comprehensively considers drug efficacy, safety and individual differences to achieve multi-dimensional screening of drugs. Drugs are categorized as therapeutic drugs for Systemic Lupus Erythematosus, Antiphospholipid Syndrome and other related complications, ensuring that recommendations meet different patient needs. Experimental results show that the Drug-ESIC-PC recommendation model we proposed has an accuracy of$\text{9 2 \%, } \text{8 8 \%}$, and$\text{8 4 \%}$for recommending two, three, and four drugs, respectively.

Drug-Drug Interactions (DDIs) are an issue that cannot be ignored in clinical treatment. With the improvement of people's living standards and the evolution of the disease spectrum, the relationship between emotions and the immune system has gradually attracted attention. Among them, mood swings can significantly affect the function of the immune system, and disorders of the immune system may also cause mood disorders. Therefore, the interaction between antidepressants and immunosuppressive drugs has important research value. This article deeply analyzes the interaction strength of these two types of drugs to assess their potential risks, thereby optimizing treatment plans and reducing adverse reactions. At the same time, 3D imaging technology is used to more intuitively display the binding position and action intensity of drugs at the target, providing strong support for clinical decision-making and helping to achieve safer and more personalized combined treatment plans.

Co-salient object detection (CoSOD) aims to detect common salient objects across multiple related images. However, existing methods often struggle with limited attention coverage, missing some co-salient objects. To address this, we propose a two-stage feature processing module (FPM) comprising comprehensive feature extraction module (CFE) and feature enhancement module (FEM). CFE extracts comprehensive cosalient features while reducing background noise, and FEM enhances feature representation and adjusts attention weights for full object coverage. Additionally, we introduce an adversarial learning module (ALM) to improve prediction quality by reducing noise in the co-salient regions. Extensive experiments on three benchmark datasets—CoCA, CoSOD3k, and CoSal2015—demonstrate that our model significantly outperforms state-of-the-art methods. The source code is available at https://github.com/yaobaimiao/CFPAM.

To address the issues of noise amplification and feature incompatibility arising from modal heterogeneity in multimodal sentiment analysis, this paper proposes a hierarchical optimization framework. In the first stage, we introduce the Semantic-Guided Calibration Network (SGC-Net), which, through a Dynamic Balancing Regulator (DBR), leverages textual semantics to intelligently weight and calibrate the cross-modal interactions of audio and video, thereby suppressing noise while preserving key dynamics. In the second stage, the Synergistic Refinement Fusion Module (SRF-Module) performs a deep refinement of the fused multi-source features. This module employs a Saliency-Gated Complementor (SGC) to rigorously filter and exchange effective information across streams, ultimately achieving feature de-redundancy and strong complementarity. Extensive experiments on the CMU-MOSI and CMU- MOSEI datasets validate the effectiveness of our method, with the model achieving state-of-the-art performance on key metrics such as binary accuracy (Acc-2: 86.73% on MOSI, 86.52% on MOSEI) and seven-class accuracy (Acc- 7: 48.35% on MOSI, 53.81% on MOSEI).

A mixed traffic environment of manual driving and automatic driving will become the norm in future intelligent transportation systems. The deep reinforcement learning (DRL) method has shown significant promise in cooperative control for traffic lights and connected autonomous vehicles (CAV) in a mixed-traffic environment. However, the uncertainty and noise in integrating agents' observations can lead to inadequate exploration of environmental data by DRL algorithms. Consequently, these algorithms are prone to overfitting and becoming trapped in local optimal, which limits the performance of control strategies. To more effectively harness the gathered environmental data and thereby facilitate improved decision-making by agents, a DRL-based cooperative control method with fuzzy feature fusion (F3DRL) was proposed in this article. First, the adaptive fuzzy inference module is implemented to adaptively mitigate information uncertainty as the data from CAV is aggregated. Then, a deep information extraction module was introduced and integrated with the output of the adaptive fuzzy inference module to establish a parallel feature fusion module. The adaptive fuzzy inference module mitigates uncertainty in the extracted traffic environmental states, while the deep information extraction module facilitates the extraction of a more comprehensive environmental representation. The fusion of features derived from these two distinct modules aids DRL agents in making better action selections, which significantly enhances the effectiveness and stability of the F3DRL method. In simulations, F3DRL significantly reduced travel and delay times, fuel consumption, and CO$_{2}$emissions, outperforming both traditional and state-of-the-art methods.

Incurable Alzheimer's disease (AD) plagues many elderly people and families. It is important to accurately diagnose and predict it at an early stage. However, the existing methods have shortcomings, such as inability to learn local and global information and the inability to extract effective features. In this paper, we propose a lightweight classification network Local and Global Graph ConvNeXt. This model has a hybrid architecture of convolutional neural network and Transformers. We build the Global NeXt Block and the Local NeXt Block to extract the local and global features of the structural magnetic resonance imaging (sMRI). These two blocks are optimized by adding global multilayer perceptron and locally grouped attention, respectively. Then, the features are fed into the pixel graph neural network to aggregate the valid pixel features using mask attention. In addition, we decoupled the loss by category to optimize the calculation of the loss. This method was tested on slices of the processed sMRI datasets from ADNI and achieved excellent performance. Our model achieves 95.81% accuracy with fewer parameters and floating point operations per second (FLOPS) than other classical efficient models in the diagnosis of AD.

Cardiovascular disease and depression often require combined use of cardiovascular and psychotropic drugs. This paper introduces Multi-Feature Fusion Learning for Discovering Synergistic Relationships (MFFL-DSR), a method to predict drug interactions. It first constructs matrices of drug features, including classification, targets, enzymes, pathways, and molecular structure. Drugs from different feature domains are then projected into a shared interaction domain. A regularization term is formulated to represent drug pairing relationships in the inter-action space, forming the MFFL-DSR objective function. Finally, iterative optimization identifies all potential drug combinations. The results show that MFFL-DSR outperforms baseline methods in six metrics: AUPR, AUC, Precision, Accuracy, Recall, and F1 score.