Current Issue : October-December Volume : 2025 Issue Number : 4 Articles : 5 Articles
Federated learning (FL) offers a privacy-preserving framework for distributed machine learning, enabling collaborative model training across diverse clients without centralizing sensitive data. However, statistical heterogeneity, characterized by non-independent and identically distributed (non-IID) client data, poses significant challenges, leading to model drift and poor generalization. This paper proposes a novel algorithm, pFedKDWCL (Personalized Federated Knowledge Distillation with Weighted Combination Loss), which integrates knowledge distillation with bi-level optimization to address non-IID challenges. pFedKD-WCL leverages the current global model as a teacher to guide local models, optimizing both global convergence and local personalization efficiently. We evaluate pFedKD-WCL on the MNIST dataset and a synthetic dataset with non-IID partitioning, using multinomial logistic regression (MLR) and multilayer perceptron models (MLP). Experimental results demonstrate that pFedKD-WCL outperforms state-of-the-art algorithms, including FedAvg, FedProx, PerFedAvg, pFedMe, and FedGKD in terms of accuracy and convergence speed. For example, on MNIST data with an extreme non-IID setting, pFedKD-WCL achieves accuracy improvements of 3.1%, 3.2%, 3.9%, 3.3%, and 0.3% for an MLP model with 50 clients compared to FedAvg, FedProx, PerFedAvg, pFedMe, and FedGKD, respectively, while gains reach 24.1%, 22.6%, 2.8%, 3.4%, and 25.3% for an MLR model with 50 clients....
Convolutional Neural Networks (CNNs) have been demonstrated to be one of the most powerful methods for image recognition, being applied in many fields, including civil and structural health monitoring in infrastructure asset management. Current State-ofthe- Art CNN models are now accessible as open-source and available on several Artificial Intelligence (AI) platforms, with TensorFlow being widely used. Besides CNN models, Vision Transformers (ViTs) have recently emerged as a competitive alternative. Several demonstrations have indicated that ViT models, in many instances, outperform the current CNNs by almost four times in terms of computational efficiency and accuracy. This paper presents an investigation into defect detection for civil and structural components using CNN and ViT models available on TensorFlow. An empirical study was conducted using a database of cracks. The severity of crack is categorized into binary states: “with crack” and “without crack”. The results confirm that the accuracies of both CNN and ViT models exceed 95% after 100 epochs of training, with no significant difference observed between them for binary classification. Notably, the cost of this AI-based approach with images taken by lightweight and low-cost drones is considerably lower compared to high-speed inspection cars, while still delivering an expected level of predictive accuracy....
To address the issue of the low precision in detecting defects in aluminum alloy weld seam digital radiography (DR) images using the current target detection algorithms, a modified algorithm named YOLOv8-ELA based on YOLOv8 is proposed. The model integrates a novel HS-FPN feature fusion module, which optimizes the parameter efficiency and enhances the detection performance. For better identification of small defect features, the CA attention mechanism within HS-FPN is substituted with the ELA attention mechanism. Additionally, the first output layer is enhanced with a SimAM attention mechanism to improve the small target recognition. The experimental findings indicate that, at a 0.5 threshold, the YOLOv8-ELA model achieves mean average precision (mAP@0.5) values of 93.3%, 96.4%, and 96.5% for detecting pores, inclusions, and incomplete welds, respectively. These values surpass those of the original YOLOv8 model by 1.4, 2.3, and 0.1 percentage points. Overall, the model attains an average mAP of 95.4%, marking a 1.3% improvement over its predecessor, confirming its superior defect detection capabilities....
Nowadays, high precision and reliability of Global Navigation Satellite Systems are increasingly important in positioning applications. Machine learning is used to improve the performance of the GSHARP PPP algorithm by reducing the effect of multipath on GNSS measurements. The clustering analysis is conducted on the primary GNSS data points with the goal of discovering and analyzing patterns in the multipath interference. This study represents an early attempt to apply AI to the GSHARP PPP algorithm. Since Lightweight Machine Learning is used in this research, it is easier to integrate and might lay the groundwork for future integration of advanced deep learning methods. About 50 h of data collected from different environments (e.g., highways and urban areas) serves as the training data for these algorithms, which ensures their robustness and real-world applicability. The use of machine learning clustering inside the PPP algorithm serves as a way to improve its performance against multipath effects, as well as provide a platform for subsequent development of precision GNSS systems through AI technologies....
Glaucoma remains a leading cause of irreversible blindness worldwide, with early detection crucial for preventing vision loss. This study developed and validated a novel eyetracking algorithm to detect oculomotor abnormalities in primary open-angle glaucoma (POAG). We conducted a case–control study (March–June 2021), recruiting 16 patients with moderate POAG, 16 with preperimetric POAG, and 16 age-matched controls. The participants underwent a comprehensive ophthalmic examination and eye movement recording using a high-resolution infrared tracker during two tasks: saccades to static targets and saccades to moving targets. The patients with POAG exhibited a significantly increased saccadic latency and reduced accuracy compared to the controls, with more pronounced differences in the moving target task. Notably, preperimetric POAG patients showed significant abnormalities despite having normal visual fields based on standard perimetry. Our machine learning algorithm incorporating multiple saccadic parameters achieved an excellent discriminative ability between glaucomatous and healthy subjects (AUC = 0.92), with particularly strong performance for moderate POAG (AUC = 0.97) and good performance for preperimetric POAG (AUC = 0.87). These findings suggest that eye movement analysis may serve as a sensitive biomarker for early glaucomatous damage, potentially enabling earlier intervention and improved visual outcomes....
Loading....