Hasil untuk "deep learning"

Meenakshi Sudarvizhi Seenipeyathevar, Prasath Palaniappan, Vijayakumar Arumugam et al.

ABSTRACT This study deals with an integrated experimental‐machine learning framework for wear estimation in functionally graded composites made from recycled magnesium machining chips, using low‐cost ceramic fibers as reinforcement with the radial Modeling technique. The primary hurdle that is being addressed is the accurate prediction of wear behavior in spatially graded magnesium matrix composites, while simultaneously avoiding extensive experimental testing. Under varying degrees of applied loads (4.4 to 39 N), sliding speeds (0.45 to 4.5 m/s), and sliding distances (500 to 4500 m), the wear performance was experimentally assessed. Results demonstrate a hardness increment of 26.26% in the outer region compared to the inner region, while resistance to wear was enhanced by 19.8% in the outer zone due to the grading of ceramic fibers. A limited experimental dataset consisting of wear measurements from the inner, middle, and outer zones of the composite was utilized in developing and validating four machine‐learning models for wear rate prediction. The tree‐based ensemble methods significantly outperformed deep‐learning strategies, with the LightGBM model providing the best prediction performance across all zones and achieving optimization with a maximum tree depth of 5, 480 leaves, and a feature fraction of 0.05. Moreover, zone‐specific XGBoost models were also developed, employing customized learning rates and minimal loss reduction parameters in order to elevate prediction accuracy. The proposed machine‐learning framework thus provides a pathway for rapid and reliable wear rate estimation for ceramic fiber‐reinforced magnesium composites, significantly lessening experimental burden. Results highlight that recycled magnesium waste, when combined with ceramic reinforcement, can be effectively employed to produce sustainable and economically viable materials with improved wear resistance, particularly for automotive and industrial applications.

Doi Thi Lan, Quan T. Ngo, Luong Vuong Nguyen et al.

Abstract In cognitive radio (CR) systems, the accurate detection of spectrum holes is a cornerstone for efficient spectrum utilization. However, the increasing complexity of CR environments, particularly those with multiple primary users (PUs), has made precise spectrum sensing a paramount challenge. To address this challenge, this study introduces the ATC model, a novel deep learning architecture that integrates a parallel combination of attention mechanism-based networks and a Convolutional Neural Network (CNN). This hybrid design enables the model to capture both spatial and temporal features from the distinct statistics of sensing signals, thereby enhancing the accuracy of spectrum state detection. The model employs a Graph Attention Network (GAT) to extract complex topological features from graph-structured data derived from received signal strength, dynamically highlighting the most relevant information. To complement this, a CNN processes the sample covariance matrix of sensing signals, unlocking localized statistical correlations and hierarchical feature representations by treating the matrix as an image. Temporal dynamics, such as PU activity patterns, are modeled using a Transformer encoder, which leverages a self-attention mechanism to learn sequential features effectively. The proposed model is evaluated using both simulated and real-world datasets. For the simulated datasets, the model is assessed and compared with baseline methods under multi-PU scenarios across different channel models. For the real-world dataset, the experimental setup is configured for a single-PU scenario due to practical data collection limitations. In both cases, the ATC model demonstrates improved performance over the benchmarked spectrum sensing methods, exhibiting higher accuracy and robustness within the respective evaluation settings.

Yong Wang, Hongwen He, Yuankai Wu et al.

An effective energy management strategy (EMS) is essential to optimize the energy efficiency of electric vehicles (EVs). With the advent of advanced machine learning techniques, the focus on developing sophisticated EMS for EVs is increasing. Here, we introduce LearningEMS: a unified framework and open-source benchmark designed to facilitate rapid development and assessment of EMS. LearningEMS is distinguished by its ability to support a variety of EV configurations, including hybrid EVs, fuel cell EVs, and plug-in EVs, offering a general platform for the development of EMS. The framework enables detailed comparisons of several EMS algorithms, encompassing imitation learning, deep reinforcement learning (RL), offline RL, model predictive control, and dynamic programming. We rigorously evaluated these algorithms across multiple perspectives: energy efficiency, consistency, adaptability, and practicability. Furthermore, we discuss state, reward, and action settings for RL in EV energy management, introduce a policy extraction and reconstruction method for learning-based EMS deployment, and conduct hardware-in-the-loop experiments. In summary, we offer a unified and comprehensive framework that comes with three distinct EV platforms, over 10  000 km of EMS policy data set, ten state-of-the-art algorithms, and over 160 benchmark tasks, along with three learning libraries. Its flexible design allows easy expansion for additional tasks and applications. The open-source algorithms, models, data sets, and deployment processes foster additional research and innovation in EV and broader engineering domains.

Ifiok Udoidiok, Fuhao Li, Jielun Zhang

Machine learning (ML) has become a cornerstone of critical applications, but its vulnerability to data poisoning attacks threatens system reliability and trustworthiness. Prior studies have begun to investigate the impact of data poisoning and proposed various defense or evaluation methods; however, most efforts remain limited to quantifying performance degradation, with little systematic comparison of internal behaviors across model architectures under attack and insufficient attention to interpretability for revealing model vulnerabilities. To tackle this issue, we build a reproducible evaluation pipeline and emphasize the importance of integrating robustness with interpretability in the design of secure and trustworthy ML systems. To be specific, we propose a unified poisoning evaluation framework that systematically compares traditional ML models, deep neural networks, and large language models under three representative attack strategies including label flipping, random corruption, and adversarial insertion, at escalating severity levels of 30%, 50%, and 75%, and integrate LIME-based explanations to trace the evolution of model reasoning. Experimental results demonstrate that traditional models collapse rapidly under label noise, whereas Bayesian LSTM hybrids and large language models maintain stronger resilience. Further interpretability analysis uncovers attribution failure patterns, such as over-reliance on neutral tokens or misinterpretation of adversarial cues, providing insights beyond accuracy metrics.

Hengjin Ke, Cang Cai, Jia Wu et al.

Hemalatha Kanakarajan, Wouter De Baene, Patrick Hanssens et al.

Abstract Background and purpose Accurate segmentation of brain metastases on Magnetic Resonance Imaging (MRI) is tedious and time-consuming for radiologists that could be optimized with deep learning (DL). Previous studies assessed several DL algorithms focusing only on training and testing the models on the planning MRI only. The purpose of this study is to evaluate well-known DL approaches (nnU-Net and MedNeXt) for their performance on both planning and follow-up MRI. Materials and methods Pre-treatment brain MRIs were retrospectively collected for 255 patients at Elisabeth-TweeSteden Hospital (ETZ): 201 for training and 54 for testing, including follow-up MRIs for the test set. To increase heterogeneity, we added the publicly available MRI scans from the Mathematical oncology laboratory of 75 patients to the training data. The performance was compared between the two models, with and without the addition of the public data. To statistically compare the Dice Similarity Coefficient (DSC) of the two models trained on different datasets over multiple time points, we used Linear Mixed Models. Results All models obtained a good DSC (DSC > = 0.93) for planning MRI. MedNeXt trained with combined data provided the best DSC for follow-ups at 6, 15, and 21 months (DSC of 0.74, 0.74, and 0.70 respectively) and jointly the best DSC for follow-ups at three months with MedNeXt trained with ETZ data only (DSC of 0.78) and 12 months with nnU-Net trained with combined data (DSC of 0.71). On the other hand, nnU-Net trained with combined data provided the best sensitivity and FNR for most follow-ups. The statistical analysis showed that MedNeXt provides higher DSC for both datasets and the addition of public data to the training dataset results in a statistically significant increase in performance in both models. Conclusion The models achieved a good performance score for planning MRI. Though the models performed less effectively for follow-ups, the addition of public data enhanced their performance, providing a viable solution to improve their efficacy for the follow-ups. These algorithms hold promise as a valuable tool for clinicians for automated segmentation of planning and follow-up MRI scans during stereotactic radiosurgery treatment planning and response evaluations, respectively. Clinical trial number Not applicable.

Xinyue Wang, Weifan Lin, Weiting Zhang et al.

In this paper, the Merkle-Transformer model is introduced as an innovative approach designed for financial data processing, which combines the data integrity verification mechanism of Merkle trees with the data processing capabilities of the Transformer model. A series of experiments on key tasks, such as financial behavior detection and stock price prediction, were conducted to validate the effectiveness of the model. The results demonstrate that the Merkle-Transformer significantly outperforms existing deep learning models (such as RoBERTa and BERT) across performance metrics, including precision, recall, accuracy, and F1 score. In particular, in the task of stock price prediction, the performance is notable, with nearly all evaluation metrics scoring above 0.9. Moreover, the performance of the model across various hardware platforms, as well as the security performance of the proposed method, were investigated. The Merkle-Transformer exhibits exceptional performance and robust data security even in resource-constrained environments across diverse hardware configurations. This research offers a new perspective, underscoring the importance of considering data security in financial data processing and confirming the superiority of integrating data verification mechanisms in deep learning models for handling financial data. The core contribution of this work is the first proposition and empirical demonstration of a financial data analysis model that fuses data integrity verification with efficient data processing, providing a novel solution for the fintech domain. It is believed that the widespread adoption and application of the Merkle-Transformer model will greatly advance innovation in the financial industry and lay a solid foundation for future research on secure financial data processing.

Gabriel de Sousa Meira, João Victor Ferreira Guedes, Edilson de Souza Bias

The use of geotechnologies in the field of diagnostic engineering has become ever more present in the identification of pathological manifestations in buildings. The implementation of Unmanned Aerial Vehicles (UAVs) and embedded sensors has stimulated the search for new data processing and validation methods, considering the magnitude of the data collected during fieldwork and the absence of specific methodologies for each type of sensor. Regarding data processing, the use of deep learning techniques has become widespread, especially for the automation of processes that involve a great amount of data. However, just as with the increasing use of embedded sensors, deep learning necessitates the development of studies, particularly those focusing on neural networks that better represent the data to be analyzed. It also requires the enhancement of practices to be used in fieldwork, especially regarding data processing. In this context, the objective of this study is to review the existing literature on the use of embedded technologies in UAVs and deep learning for the identification and characterization of pathological manifestations present in building façades in order to develop a robust knowledge base that is capable of contributing to new investigations in this field of research.

Shuai Wang, Zongbao Zhang, Chao Wang

Abstract The mining of open pit mines is widespread in China, and there are many cases of landslide accidents. Therefore, the problem of slope stability is highlighted. The stability of the slope is a factor that directly affects the mining efficiency and the safety of the entire mining process. According to the statistics, there is a 15 percent chance of finding landslide risk in China’s large-scale mines. And due to the expansion of the mining scale of the enterprise, the problem of slope stability has become increasingly obvious, which has become a major subject in the study of open-pit mine engineering. In order to better predict the slope stability coefficient, this study takes a mine in China as a case to deeply discuss the accuracy of different algorithms in the stability calculation, and then uses a deep learning algorithm to study the stability under rainfall conditions. The change of the coefficient and the change of the stability coefficient before and after the slope treatment are experimentally studied with the displacement of the monitoring point. The result shows that the safety coefficient calculated by the algorithm in this paper is about 7% lower than that of the traditional algorithm. In the slope stability analysis before treatment, the safety factor calculated by the algorithm in this paper is 1.086, and the algorithm in this paper is closer to reality. In the stability analysis of the slope after treatment, the safety factor calculated by the algorithm in this paper is 1.227, and the stability factor meets the requirements of the specification. It also shows that the deep learning algorithm effectively improves the efficiency of the slope stability factor prediction and improves security during project development.

Michela Prunella, Roberto Maria Scardigno, Domenico Buongiorno et al.

Automatic vision-based inspection systems have played a key role in product quality assessment for decades through the segmentation, detection, and classification of defects. Historically, machine learning frameworks, based on hand-crafted feature extraction, selection, and validation, counted on a combined approach of parameterized image processing algorithms and explicated human knowledge. The outstanding performance of deep learning (DL) for vision systems, in automatically discovering a feature representation suitable for the corresponding task, has exponentially increased the number of scientific articles and commercial products aiming at industrial quality assessment. In such a context, this article reviews more than 220 relevant articles from the related literature published until February 2023, covering the recent consolidation and advances in the field of fully-automatic DL-based surface defects inspection systems, deployed in various industrial applications. The analyzed papers have been classified according to a bi-dimensional taxonomy, that considers both the specific defect recognition task and the employed learning paradigm. The dependency on large and high-quality labeled datasets and the different neural architectures employed to achieve an overall perception of both well-visible and subtle defects, through the supervision of fine or/and coarse data annotations have been assessed. The results of our analysis highlight a growing research interest in defect representation power enrichment, especially by transferring pre-trained layers to an optimized network and by explaining the network decisions to suggest trustworthy retention or rejection of the products being evaluated.

Esraa Khatab, Ahmed Onsy, Ahmed Abouelfarag

One of the primary tasks undertaken by autonomous vehicles (AVs) is object detection, which comes ahead of object tracking, trajectory estimation, and collision avoidance. Vulnerable road objects (e.g., pedestrians, cyclists, etc.) pose a greater challenge to the reliability of object detection operations due to their continuously changing behavior. The majority of commercially available AVs, and research into them, depends on employing expensive sensors. However, this hinders the development of further research on the operations of AVs. In this paper, therefore, we focus on the use of a lower-cost single-beam LiDAR in addition to a monocular camera to achieve multiple 3D vulnerable object detection in real driving scenarios, all the while maintaining real-time performance. This research also addresses the problems faced during object detection, such as the complex interaction between objects where occlusion and truncation occur, and the dynamic changes in the perspective and scale of bounding boxes. The video-processing module works upon a deep-learning detector (YOLOv3), while the LiDAR measurements are pre-processed and grouped into clusters. The output of the proposed system is objects classification and localization by having bounding boxes accompanied by a third depth dimension acquired by the LiDAR. Real-time tests show that the system can efficiently detect the 3D location of vulnerable objects in real-time scenarios.

Nikhil Ketkar, Jojo Moolayil

Kamel Arafet, Rafael Berlanga

The generation of electricity through renewable energy sources increases every day, with solar energy being one of the fastest-growing. The emergence of information technologies such as Digital Twins (DT) in the field of the Internet of Things and Industry 4.0 allows a substantial development in automatic diagnostic systems. The objective of this work is to obtain the DT of a Photovoltaic Solar Farm (PVSF) with a deep-learning (DL) approach. To build such a DT, sensor-based time series are properly analyzed and processed. The resulting data are used to train a DL model (e.g., autoencoders) in order to detect anomalies of the physical system in its DT. Results show a reconstruction error around 0.1, a recall score of 0.92 and an Area Under Curve (AUC) of 0.97. Therefore, this paper demonstrates that the DT can reproduce the behavior as well as detect efficiently anomalies of the physical system.

Jiuxin Wang, Yutian Luo, Zhengming Yang et al.

In order to improve the measurement speed and prediction accuracy of unconventional reservoir parameters, the deep neural network (DNN) is used to predict movable fluid percentage of unconventional reservoirs. The Adam optimizer is used in the DNN model to ensure the stability and accuracy of the model in the gradient descent process, and the prediction effect is compared with the back propagation neural network (BPNN), K-nearest neighbor (KNN), and support vector regression model (SVR). During network training, L₂ regularization is used to avoid over-fitting and improve the generalization ability of the model. Taking nuclear magnetic resonance (NMR) T₂ spectrum data of laboratory unconventional core as input features, the influence of model hyperparameters on the prediction accuracy of reservoir movable fluids is also experimentally analyzed. Experimental results show that, compared with BPNN, KNN, and SVR, the deep neural network model has a better prediction effect on movable fluid percentage of unconventional reservoirs; when the model depth is five layers, the prediction accuracy of movable fluid percentage reaches the highest value, the predicted value of the DNN model is in high agreement with the laboratory measured value. Therefore, the movable fluid percentage prediction model of unconventional oil reservoirs based on the deep neural network model can provide certain guidance for the intelligent development of the laboratory’s reservoir parameter measurement.

Jiying Han, Hongbiao Yin, Xin Yang et al.

Following Grandey's integrative model of emotional labor and emotion regulation, this study examined the relationships between university teachers' reported use of various emotional labor strategies and some antecedents (i. e., perceived emotional job demands and teaching support) and teaching efficacy. A sample of 643 university teachers from 50 public higher education institutions in an East China province responded to a questionnaire survey. The data analysis based on descriptive statistics and structural equation modeling showed that surface acting impeded teaching efficacy in instructional strategy and learning assessment, while deep acting and expression of naturally felt emotions enhanced teaching efficacy in course design, instructional strategy, and learning assessment. For the antecedents of university teachers' emotional labor strategies, teachers perceived that the emotional job demands of teaching facilitated their use of surface and deep acting; in contrast, teachers' perceived teaching support decreased their use of surface acting and increased their use of expression of naturally felt emotions.

Natasha Dropka, Stefan Ecklebe, Martin Holena

The aim of this study was to assess the aptitude of the recurrent Long Short-Term Memory (LSTM) neural networks for fast and accurate predictions of process dynamics in vertical-gradient-freeze growth of gallium arsenide crystals (VGF-GaAs) using datasets generated by numerical transient simulations. Real time predictions of the temperatures and solid–liquid interface position in GaAs are crucial for control applications and for process visualization, i.e., for generation of digital twins. In the reported study, an LSTM network was trained on 1950 datasets with 2 external inputs and 6 outputs. Based on network performance criteria and training results, LSTMs showed the very accurate predictions of the VGF-GaAs growth process with median root-mean-square-error (RMSE) values of 2 × 10⁻³. This deep learning method achieved a superior predictive accuracy and timeliness compared with more traditional Nonlinear AutoRegressive eXogenous (NARX) recurrent networks.