Neural Computing and Applications最新文献

Radiological findings play an essential and complementary role in diagnosing Covid-19, assessing its severity, and managing its patients. Artificial intelligence technology based on medical imaging, which has made exciting developments by being applied in many areas, has become an area of interest for the rapid and accurate detection of the disease in the fight against the Covid-19 pandemic. The main difficulty is the inability to obtain a large dataset size with quality and standard images that neural networks need to perform well. Aiming at this problem, this study proposes a Siamese neural network-based deep learning framework for accurate diagnostics of Covid-19 using chest X-ray (CXR) images. The pre-trained VGG16 architecture, based on the transfer learning approach, forms the backbone of the Siamese neural network. The outputs of the backbones are joined together by a merging layer, and then the output passes through a fully connected layer. Based on this structure, category-aware Siamese-based models are produced for each class. The predictions from the models are combined using a voting mechanism to reduce the possibility of misclassification and to make better decisions. The framework was evaluated using a publicly available dataset for the 4-class classification task for Covid-19 pneumonia, lung opacity, normal, and non-Covid-19 viral pneumonia images. The findings reveal the high discrimination ability of the framework, trained using only 10 images per class in less training time, achieving an average test accuracy of 92%. Our framework, which learns a single Siamese-based pairwise model for each class, effectively captures class-specific features. Additionally, it has the potential to deal with data scarcity and long training time problems in multi-class classification tasks.

Coronary Artery Bypass Graft (CABG) is a common cardiac surgery, but it continues to have many associated risks, including the need for blood transfusions. Previous research has shown that blood transfusion during CABG surgery is associated with an increased risk for infection and mortality. The current study aims to use modern techniques, such as deep neural networks and data synthesis, to develop models that can best predict the need for blood transfusion among CABG patients. Results show that neural networks with synthetic data generated by DataSynthesizer have the best performance. Implications of results and future directions are discussed.

The design of a helicopter is an intricate and challenging process. Decisions made during the preliminary design phase can significantly impact subsequent design stages, making it crucial to base these decisions on a solid foundation. A range of methods, including hand calculations, finite element analyses, and experimental tests, can be employed to establish the conceptual design parameters. However, these methods often come with the drawbacks of being time-intensive and costly, especially when testing various structures during the early design phase. To address this issue, this study introduces an artificial neural network-based design tool to evaluate the static structural characteristics of a helicopter’s horizontal stabilizer. The tool was built in Python using the Keras library. The required database for the training of the artificial neural network model was established using finite element analyses of the horizontal stabilizer subjected to the aerodynamic load for diverse design variables. The model’s performance was evaluated, and the model’s outputs were compared to the results derived from the finite element analyses. Moreover, the Hammersley sampling methodology was employed to reduce the size of the database without compromising on accuracy. The study also assessed the impact of decreasing the amount of data fed into the network model.

Arithmetic optimization algorithm (AOA) is a recently proposed algorithm inspired by mathematical operations. It has been used to solve a variety of optimization problems due to its simplicity of parameters and ease of implementation. However, it has been found that AOA encounters challenges such as poor exploration and premature convergence. To solve these issues, this paper proposes a self-adaptive AOA with hybrid search modes, named AOAHSM. In this algorithm, two hybrid search modes, i.e., the parallel search mode and the serial search mode, are established by combining AOA and differential evolution (DE) in different ways to enhance the exploration and exploitation abilities, respectively. In the parallel search mode, AOA and DE independently implement on their respective subpopulations to maintain a high distribution of the population. In the serial search mode, DE is embedded into AOA to provide more diversified solutions and thereby help the population jump out of local optima. Then, a self-adaptive conversion strategy is employed to dynamically switch between the two modes so as to achieve a better balance between exploration and exploitation. Additionally, a Levy flight strategy is used to perturb and update the best solution obtained in each iteration to further prevent premature convergence. Lastly, a binary version of AOAHSM is proposed to tackle the 0–1 knapsack problem. The proposed algorithms are evaluated on CEC2019, CEC2020 test functions, two typical engineering design problems and 45 instances of the 0–1 knapsack problem and compared with a number of state-of-the-art meta-heuristic algorithms. The obtained results demonstrate that AOAHSM and its binary version not only significantly outperform the original AOA but also achieve superior performance to the comparison algorithms in most cases.

Osteoporosis (OP) is the most prevalent and common bone disease, especially knee osteoporosis. It significantly disables sufferers all over the world. Although laborious and prone to user variation, manual diagnosis, segmentation, and annotation of knee joints continue to be the preferred way to diagnose OP in clinical procedures. Therefore, many deep learning algorithms, particularly the convolutional neural network (CNN), have been created to increase clinical workflow efficiency to overcome the shortcomings of the widely used method as above. Medical imaging procedures can show hidden structures in a volumetric view, particularly those that generate three-dimensional (3D) pictures like MRI. We created a dataset of 240 pictures from patients who had knee X-rays and skeletal bone mineral density assessments at the same time. Four convolutional neural networks (CNN) models were used to analyse the X-ray images and deep neural networks for clinical covariances to determine the degree of osteoporosis. Additionally, we investigated ensemble models that included each CNN with a clinical covariance. For every network, scores for accuracy and error rate were computed. ResNet and Alexnet displayed the highest levels of accuracy when the CNN models were tested using knee X-rays with normal, low BMD, and osteoporosis. An ensemble of DNN with Alexnet, ResNet, and both ResNet and Alexnet are employed resulting in improved accuracy. The ensemble of best-performing CNN and DNN is proposed to diagnose osteoporosis more accurately. The proposed method has produced a highly accurate osteoporosis diagnosis.

Ensuring environmental safety and regulatory compliance at Department of Energy (DOE) sites demands an efficient and reliable detection system for low-level nuclear waste (LLW). Unlike existing methods that rely on human effort, this paper explores the integration of computer vision algorithms to automate the identification of such waste across DOE facilities. We evaluate the effectiveness of multiple algorithms in classifying nuclear waste materials and their adaptability to newly emerging LLW. Our research introduces and implements five state-of-the-art computer vision models, each representing a different approach to the problem. Through rigorous experimentation and validation, we evaluate these algorithms based on performance, speed, and adaptability. The results reveal a noteworthy trade-off between detection performance and adaptability. YOLOv7 shows the best performance and requires the highest effort to detect new LLW. Conversely, OWL-ViT has lower performance than YOLOv7 and requires minimal effort to detect new LLW. The inference speed does not strongly correlate with performance or adaptability. These findings offer valuable insights into the strengths and limitations of current computer vision algorithms for LLW detection. Each developed model provides a specialized solution with distinct advantages and disadvantages, empowering DOE stakeholders to select the algorithm that aligns best with their specific needs.

Power electronic converters play a crucial role in integrating distributed generation, renewable energy sources, microgrids, and HVDC transmission networks into the grid. The control technique used in the voltage source inverters (VSI) is essential for handling load variations, system nonlinearity, stability, and fast transient response. This study focuses on improving the robustness and control performance of VSIs by integrating a Kalman filter adaptive observer into a finite control set model predictive control (FCS-MPC), resulting in an improved FCS-MPC strategy (IMPC). The classical FCS-MPC can be affected by inaccuracies due to measurement noise and uncertainties in system models, leading to less accurate predictions and suboptimal control actions. By employing the Kalman filter adaptive observer, real-time estimates of unmeasured variables are provided, compensating for uncertainties, and enhancing control performance. To further enhance flexibility and adaptivity, an artificial neural network (ANN)-based controller is designed. The ANN controller is trained offline using IMPC as baseline thus eliminating the need for online predictions and optimization. The ANN controller directly generates inverter switching configuration states, resulting in high-quality sinusoidal output voltage with low distortions. Comparative analysis is conducted for the classical FCS-MPC, IMPC, support vector machine (SVM), convolutional neural network (CNN), and ANN-based controllers under diverse operating conditions and system parameters. Although it has reduced interpretability, the ANN controller exhibits superior harmonic reduction, outperforming both MPC-based controllers and SVM. Evaluation against CNN-based controls also validates the ANN’s robustness and effectiveness in handling uncertainties, emphasizing its adaptability, efficiency, and practical applicability in power electronic applications.

Deep learning models are often evaluated in scenarios where the data distribution is different from those used in the training and validation phases. The discrepancy presents a challenge for accurately predicting the performance of models once deployed on the target distribution. Domain adaptation and generalization are widely recognized as effective strategies for addressing such shifts, thereby ensuring reliable performance. The recent promising results in applying vision transformers in computer vision tasks, coupled with advancements in self-attention mechanisms, have demonstrated their significant potential for robustness and generalization in handling distribution shifts. Motivated by the increased interest from the research community, our paper investigates the deployment of vision transformers in domain adaptation and domain generalization scenarios. For domain adaptation methods, we categorize research into feature-level, instance-level, model-level adaptations, and hybrid approaches, along with other categorizations with respect to diverse strategies for enhancing domain adaptation. Similarly, for domain generalization, we categorize research into multi-domain learning, meta-learning, regularization techniques, and data augmentation strategies. We further classify diverse strategies in research, underscoring the various approaches researchers have taken to address distribution shifts by integrating vision transformers. The inclusion of comprehensive tables summarizing these categories is a distinct feature of our work, offering valuable insights for researchers. These findings highlight the versatility of vision transformers in managing distribution shifts, crucial for real-world applications, especially in critical safety and decision-making scenarios.

The rising incidence of cancer underscores the imperative for innovative diagnostic and prognostic methodologies. This study delves into the potential of RNA-Seq gene expression data to enhance cancer classification accuracy. Introducing a pioneering approach, we model gene expression data as point clouds, capitalizing on the data's intrinsic properties to bolster classification performance. Utilizing PointNet, a typical technique for processing point cloud data, as our framework's cornerstone, we incorporate inductive biases pertinent to gene expression and pathways. This integration markedly elevates model efficacy, culminating in developing an end-to-end deep learning classifier with an accuracy rate surpassing 99%. Our findings not only illuminate the capabilities of AI-driven models in the realm of oncology but also highlight the criticality of acknowledging biological dataset nuances in model design. This research provides insights into application of deep learning in medical science, setting the stage for further innovation in cancer classification through sophisticated biological data analysis. The source code for our study is accessible at: https://github.com/cialab/GPNet.

As internet information technology continues to advance, individuals are increasingly encountering and managing a vast volume of data and information. A large and complex amount of information hinders the effective transmission of valuable information, making it difficult to find multimedia content of interest in the vastness of the internet. As the volume of multimedia content rapidly grows, personalized recommendation algorithms play a crucial role in matching relevant content to users. Knowledge graphs, due to their powerful organizational and relationship processing capabilities, are commonly applied in intelligent search engines and recommendation systems. This article focuses on the effective utilization of semantic association information in knowledge graphs for multimedia content recommendation. Two main areas of research are conducted. In this article, two novel approaches are presented. To begin with, the primary objective is to improve the learning of knowledge feature representation. This is achieved by introducing a model based on self-attention, which effectively captures the diverse significance of triplets in determining the semantics of entities. This leads to improved quality of knowledge feature representation, thereby serving as valuable auxiliary information for multimedia content recommendation systems. Secondly, the article addresses the integration of knowledge graphs in multimedia content recommendation applications. This paper proposes a content recommendation algorithm that integrates a combined embedding of behavior and knowledge features. By leveraging past preferences and utilizing the semantic structure of knowledge graphs, this algorithm provides a comprehensive exploration of user interests and hobbies. Ultimately, this conducts extensive experiments to assess the effectiveness and performance of the proposed algorithms. The results validate the feasibility and efficacy of these algorithms in enhancing multimedia content recommendation systems.