Computational Intelligence最新文献

Personalized federated learning represents a pivotal strategy for addressing the challenges posed by statistical heterogeneity in federated learning. Clients optimize their models by leveraging information from other clients through a global model. However, data heterogeneity constrains the generalization capacity of the global model, thereby degrading the feature representation capability of local client models, especially in clients with limited data. In response to this challenge, we propose the Federal Merging Subgroup Information (FedMSI) method to augment personalized information in personalized federated learning. On the server side, FedMSI employs model clustering to identify subgroups of clients with similar personalized data distributions. It then aggregates cluster center models within each subgroup and transmits them to clients for use in the subsequent round of assisted training. On the client side, FedMSI introduces a local optimization objective that incorporates the cluster center model, enabling the extraction of informative knowledge to enhance local training. Experiments demonstrate the effectiveness of FedMSI across different datasets, data heterogeneity levels, and data sizes. Ablation experiments further confirm the effectiveness of the design of the local optimization objective. Compared to state-of-the-art methods, FedMSI achieves a 13.16% improvement in scalability performance accuracy.

As modelling of nonlinear oscillator systems plays an important part in science and engineering fields, a double loop Radial Basis Function Neural Network (RBFNN) with adaptive radius and lattices is proposed for handling this issue. In this design, a large enough lattice arranged to cover all of the trajectories is taken as the mapping center of the RBFNN at the initial condition. The number of lattices will be dynamically adjusted, and those lattices far from the trajectories will be removed. Applying Taylor expansion in local space, the activated radius factor can be separated from the Gaussian function. In order to guarantee that the modelling scheme has the characteristic of fast convergence, the error power function is utilized to minimize the gain parameter of the error differential equation. In the double loop structure, the updated equation of weights and activated radius can be determined by the Lyapunov function, which can guarantee that the weights and the activated radius will converge to the neighborhood of their true value and the tracking error of state trajectories will converge to the neighborhood of zero. In order to show the effectiveness and superiority of the double loop RBFNN proposed in this paper, Helmholtz–Duffing and Vanderpol–Duffing are used as the testing objects of the nonlinear oscillator system while comparing with Deterministic Learning.

Aspect-based Sentiment Analysis (ABSA) is a vital NLP task that identifies sentiment towards specific entities or aspect terms within a text. Recently, large language models (LLMs) have shown impressive capabilities in semantic comprehension and logical inference. However, LLM hallucinations pose challenges in accurately determining sentiment polarity for aspect terms, leading to performance issues. Moreover, current ABSA methods often fail to fully leverage the vast prior knowledge embedded within LLMs, resulting in suboptimal classification outcomes for specific aspects. Inspired by these challenges, we propose the BYD-OBS-ABSA framework—‘Beyond Simple Observations, Embracing Comprehensive Contextual Insights’ for ABSA tasks. This framework leverages unique in-context constraints, backgrounds, and analogical reasoning to address LLM hallucinations and uses self-adaptive bootstrap instructions optimization to enhance LLM predictions. BYD-OBS-ABSA integrates various in-context augmentation strategies, including emotion-oriented backgrounds, constraints, and analogical reasoning. BYD-OBS-ABSA further improves initial LLM instructions through adaptive iterative optimization using a random search bootstrap algorithm, maximizing the benefits of LLM prompting. Extensive zero/few-shot experiments with GPT-3.5-turbo across six public datasets validate the effectiveness and robustness of our framework, even surpassing human judgment in certain scenarios.

Clinicians can determine cardiac symptoms with the aid of heart rate detection and continuous surveillance. One of the necessary indicators utilized to assess the physiological wellness of the human body is heartbeat rate. Conventional heartbeat rate monitoring systems necessitate skin contact; nevertheless, Remote Photoplethysmography (RPPG) allows for contactless heartbeat rate measurement by using a video camera that records minute changes in skin tone. The medical field has increased interest in the surveillance of physiological signals due to technological developments in future healthcare. In this paper, a novel heartbeat rate monitoring system is developed to significantly track heart functioning. Initially, the facial videos are gathered from the online resources, and they are given to the face detection phase for generating detected face images with the help of Vision Transformer-based You Only Look Once v5 (ViT-Yolov5). Further, the detected face images are given to the feature extraction process. Here, the 3D Residual Attention Network (3D-RAN) is used to retrieve the features to generate the feature set 1. Similarly, Red, Green, and Blue (RGB) features and Deep Belief Network (DBN)-based features are retrieved from the input video, which is considered as feature set 2. Further, the heartbeat monitoring is done using the Weighted feature fusion-based Adaptive Long Short-Term Memory with Attention Mechanism (WALSTM-AM), and the weighted feature fusion is carried out on the feature sets 1 and 2, in which the weight and parameters are optimized using Enhanced Shell Game Optimization (ESGO). The numerical results showed that the developed model attained a Normalized Mean Square Error Value (NMSE) of 0.001868 that indicates better and more reliable validation of heartbeat rate, leading to higher representation of heart's activity. Thus, the proposed heartbeat monitoring mechanism attains better outcomes than the conventional methods to prove its optimal performance.

Falls among the elderly remain a critical public health concern, often leading to severe injuries or fatalities. In response, we propose SmartFallNet, an advanced deep learning framework designed for accurate and real-time fall detection in elderly care settings. The architecture leverages a dual-stream design that independently processes multimodal data from both RGB and depth frames using Vision Transformers (ViTs) for spatial feature encoding and Gated Recurrent Units (GRUs) for temporal sequence modeling. To enhance the model's sensitivity to fall-relevant patterns, we introduce a novel Dynamic Kernel Attention Mechanism (DKAM) that selectively emphasizes critical temporal frames. Features from both modalities are fused via a Power-Weighted Aggregation strategy, resulting in a rich and context-aware representation of human motion. Evaluated on two benchmark datasets—UR Fall Detection (URFD) and the Falling Detection Dataset (FDD)—SmartFallNet achieved state-of-the-art performance with 99.1% accuracy on URFD and 98.5% on FDD, while maintaining a low inference latency of 0.2 s per sample, supporting near real-time application. Extensive experiments, including ablation studies, demonstrate the model's robustness, efficiency, and superiority over existing methods. These results underscore the potential of SmartFallNet for deployment in real-world healthcare and ambient-assisted living environments.

Diabetic retinopathy (DR) is a common complication of type 2 diabetes. It occurs when high blood sugar levels damage the blood vessels in the retina, the light-sensitive tissue at the back of the eye. While diabetic retinopathy can occur in both type 1 and type 2 diabetes, it is indeed more commonly associated with type 2 diabetes due to its higher prevalence and longer duration in many cases. Type 2 diabetes often develops gradually over time, allowing for prolonged exposure to elevated blood sugar levels. This prolonged exposure increases the risk of developing diabetic retinopathy and other diabetes-related complications. The aim of this paper is to analyze the various deep learning models for effective prediction of diabetic retinopathy in patients suffering from Type 2 Diabetes. Furthermore, standard datasets consisting of 38,788 training and 55,504 test images for diabetic retinopathy and blindness are obtained. On the other hand, deep learning models such as ResNet101V2, DenseNet201, InceptionResNetV2, EfficientNetB7, and Xception CNNs are applied to the dataset and trained as well. Moreover, the performance of all the models is assessed on the basis of certain quality measures, such as accuracy, F1 score, recall, precision, RMSE values, and loss. On the other hand, results indicate the potential of deep learning models in accurately predicting diabetic retinopathy, thereby aiding in early diagnosis and intervention to prevent vision loss in patients with Type 2 Diabetes.

Cardiovascular diseases are the leading cause of morbidity and mortality worldwide, which requires accurate and reliable diagnostic methods. This paper introduces an effective anomaly detection framework, t-SNE-LOF, which leverages the complementary strengths of t-distributed Stochastic Neighbor Embedding (t-SNE) and Local Outlier Factor (LOF) detection. t-SNE captures complex, nonlinear relationships within high-dimensional data by reducing it to a lower-dimensional manifold while preserving local and global structures, whereas LOF identifies anomalies based on density variations, making it well-suited for detecting outliers in heterogeneous datasets. By combining these techniques, t-SNE-LOF effectively isolates cardiovascular anomalies with enhanced precision, achieving Area Under the Curve (AUC) scores of 99.28% and 98.75% on two publicly available datasets, outperforming state-of-the-art baseline methods. Comparative evaluation against a broad range of classical semi-supervised methods (iForest, LOF, EE), three embedding-based variants, seven traditional supervised classifiers, and six deep learning models confirms the effectiveness and robustness of the approach. Additionally, SHapley Additive exPlanations (SHAP) analysis interprets the model's predictions, providing insights into feature contributions and enhancing the explainability of the t-SNE-LOF framework for cardiovascular anomaly detection. The proposed method is semi-supervised and requires only healthy data during training, making it highly suitable for real-world clinical settings where labeled abnormal cases are limited. Future work will explore perplexity-free parametric t-SNE to improve scalability and eliminate manual tuning, along with automated hyperparameter optimization and extension to multi-modal cardiovascular data. This approach offers satisfactory performance and may support integration into Clinical Decision Support Systems to aid early cardiovascular diagnosis and prevention.

An important advancement in medical diagnostics is the detection of eye diseases through image analysis, which is facilitated by the development of robust machine-learning approaches. This research provided an extensive methodology for eye disease classification, introducing the Multi-Layer DenseNet-77 architecture for analyzing the images. The process starts with the aggregation of different datasets for various eye conditions, which include healthy eyes, cataracts, diabetic retinopathy, and glaucoma. The data undergoes pre-processing, including grayscale conversion, resizing, and standardization, to improve image quality and ensure uniformity. Feature extraction is performed to capture important characteristics of the images. First-order statistics, such as entropy, mean, intensity, and energy, are used for feature extraction. These features serve as inputs to the Multilayer DenseNet-77 model. To enhance the classification accuracy, the Multilayer DenseNet-77 model incorporates advanced regularization techniques and optimized training strategies. The architecture of the model facilitates efficient feature reuse through the dense connections, which tackle the challenges associated with gradient flow and overfitting. Through the pooling layers and multiple convolutional layers, the retinal images are processed, and the DenseNet-77 approach classifies the images into predefined categories that significantly improve the early diagnosis and detection of eye diseases. This method not only contributes to enhancing the outcomes of the patients but also sets the stage for using deep learning in the analysis of medical images, highlighting its potential in transforming healthcare interventions. The experimental result validates that the proposed Multilayer DenseNet-77 model attained an accuracy of 99.21%, precision of 98.65%, recall of 98.56%, F1-score of 97.42%, specificity of 98.61%, and dice coefficient of 0.974. The outcomes of the Multilayer DenseNet-77 model indicate that the model achieved excellent performance in eye disease classification.

Human prakriti—a core Ayurvedic construct—encompasses constitution-based guidance in the Vatt, Pitt, and Kaph types. While clinically helpful, prakriti assessment is highly reliant on subjectivity, limiting reproducibility and cross-cultural generality. Here, well-validated psychometric assessments are integrated with interpretable machine learning to yield objective, scalable classification. Emotional quotient (EQ), risk-taking, personality, and Raven's progressive matrices (RPM IQ) tests were given to Indian (N = 202; 76% male) and U.S. (N = 204; 53% male) samples. Seven models were tested, including logistic regression, decision tree, random forest, SVM, XGBoost, CatBoost, and multi-layer perceptron, and hand-tuned stacked ensemble (MR-CAT). Cross-cultural validation—training Indian data and validation on U.S. data—produced MR-CAT to achieve accuracies of 0.97 (India) and 0.99 (U.S.) with high precision, recall, and F1. SHapley additive explanations (SHAP) revealed differential psychometric patterns: for example, Vatt with high resilience and self-motivation and low collaborative leadership, Pitt with high calm and ethical courage, and Kaph with high emotional stability but low leadership initiative. These patterns conform to, yet go beyond, traditional dosha portraits. The findings demonstrate a reproducible, cross-culturally stable interface between Ayurvedic theory and computational intelligence, opening the door to scalable personalized preventive care and subsequent mind–body research.

Smart agriculture (SA) is referred to as an improved approach in agricultural production. In SA, particularly smart irrigation systems, the integration of Internet of Things (IoT) devices has improved automation and efficiency. The SA idea is developed by integrating Information Technology (IT) and data processing techniques with quantitative decision-making and intelligent control measures to advance production. Heterogeneous sensors are implemented in SA communication for enhanced security. However, the deployment of heterogeneous IoT sensors in large-scale networks increases the vulnerability to cyber intrusions, which can severely disrupt irrigation control and crop health. For secure SA farming, this paper proposes a SpinalFuzzyGoogLeNet (SpinalFGNet) module for intrusion detection in smart irrigation. Initially, the IoT system is simulated; the process of intrusion detection is performed by acquiring input data from specific datasets. Here, data pre-processing is done with Quantile Normalization. Then, feature fusion is achieved by Deep Kronecker Network with Gower distance. Afterward, data augmentation is performed by deploying oversampling based on Synthetic Minority Oversampling Technique. At last, intrusion detection is performed using the proposed SpinalFGNet, which is developed by the integration of SpinalNet and GoogLeNet with fuzzy concept. The evaluation of SpinalFGNet is conducted and accomplishments in terms of metrics gained accuracy at 92%, precision as 93%, recall as 95%, and F1-score as 94%.