Search

Brain–computer interfacing (BCI) research, fueled by deep learning, integrates insights from diverse domains. A notable focus is on steady-state visual evoked potential (SSVEP) in BCI applications, requiring in-depth assessment through deep learning. EEG research frequently employs SSVEPs, which are regarded as normal brain responses to visual stimuli, particularly in investigations of visual perception and attention. This paper tries to give an in-depth analysis of the implications of deep learning for SSVEP-adapted BCI. A systematic search across four stable databases (Web of Science, PubMed, ScienceDirect, and IEEE) was developed to assemble a vast reservoir of relevant theoretical and scientific knowledge. A comprehensive search yielded 177 papers that appeared between 2010 and 2023. Thence a strict screening method from predetermined inclusion criteria finally generated 39 records. These selected works were the basis of the study, presenting alternate views, obstacles, limitations and interesting ideas. By providing a systematic presentation of the material, it has made a key scholarly contribution. It focuses on the technical aspects of SSVEP-based BCI, EEG technologies and complex applications of deep learning technology in these areas. The study delivers more penetrating reporting on the latest deep learning pattern recognition techniques than its predecessors, together with progress in data acquisition and recording means suitable for SSVEP-based BCI devices. Especially in the realms of deep learning technology orchestration, pattern recognition techniques, and EEG data collection, it has effectively closed four important research gaps. To increase the accessibility of this critical material, the results of the study take the form of easy-to-read tables just generated. Applying deep learning techniques in SSVEP-based BCI applications, as the research shows, also has its downsides. The study concludes that a radical framework will be presented which, includes intelligent decision-making tools for evaluation and benchmarking. Rather than just finding a comparable or similar analogy, this framework is intended to help guide future research and pragmatic applications, and to determine which SSVEP-based BCI applications have succeeded at responsibility for what they set out with. [ABSTRACT FROM AUTHOR]

In recent years, deep learning has significantly reshaped numerous fields and applications, fundamentally altering how we tackle a variety of challenges. Areas such as natural language processing (NLP), computer vision, healthcare, network security, wide-area surveillance, and precision agriculture have leveraged the merits of the deep learning era. Particularly, deep learning has significantly improved the analysis of remote sensing images, with a continuous increase in the number of researchers and contributions to the field. The high impact of deep learning development is complemented by rapid advancements and the availability of data from a variety of sensors, including high-resolution RGB, thermal, LiDAR, and multi-/hyperspectral cameras, as well as emerging sensing platforms such as satellites and aerial vehicles that can be captured by multi-temporal, multi-sensor, and sensing devices with a wider view. This study aims to present an extensive survey that encapsulates widely used deep learning strategies for tackling image classification challenges in remote sensing. It encompasses an exploration of remote sensing imaging platforms, sensor varieties, practical applications, and prospective developments in the field. [ABSTRACT FROM AUTHOR]

Timely estimation of earthquake magnitude plays a crucial role in the early warning systems for earthquakes. Despite the inherent danger associated with earthquake energy, earthquake research necessitates extensive parameter estimation and predictive techniques to account for uncertain trends in earthquake waveforms when determining earthquake magnitudes using a single station. This study introduces an effective solution to tackle the issue through the automatic magnitude deep network (AMagDN) model. The proposed model includes long short-term memory (LSTM), a bidirectional LSTM, an autocorrelation attention mechanism, and a machine learning block that can capture detailed information from the seismic waveform recorded during an earthquake. The unique feature of this model is the use of multivariate time series waveforms derived from recorded accelerograms specifically tailored to their energy significance with magnitude and seven fusion tabular parameters involving source and geospatial features. The proposed model's training, validation and testing are done using independent 15014, 1287 and 3448 records maintained by the Kyoshin network, Japan, for moderate to great impact earthquakes between 5.5 and 8.0 ( M JMA ). A comparative study shows that the proposed model outperforms recent state-of-the-art models and common linear relations, reducing mean absolute prediction error by 40% from the second-best model. The multi-stations data are also used for successfully forecasting the magnitudes of two significant earthquakes of 7.7 and 7.3 magnitude ( M JMA ) using the proposed model. The reliable prediction capabilities of the proposed model for both single and multi-station data clearly demonstrate its utility in reducing earthquake hazards. [ABSTRACT FROM AUTHOR]

The demand for accurate health status assessment of bearings in rotating electrical machines is rising. However, traditional fault diagnosis methods based on deep learning are not effective in efficiently and intelligently determining the health condition of the bearings and also requires large memory, hence constraining their use in low-cost microcontrollers. In this work, combination of short time Fourier transform (STFT) with Ostu's threshold method and MobileNet-V2 with U-Net (MV2UNet) is proposed for rolling bearing fault health status. The proposed method obtains the STFT images of the raw vibration signals from the laboratory experimental setup. Then the Ostu's threshold method is used to make the labels for STFT images. Finally, the proposed method is trained with these labelled images to detect the health status of each bearing fault. The proposed model achieved the Dice coefficient is 0.9834 and the intersection over union is 0.9479. The main advantage of the proposed model has low memory requirements, which makes it well-suited for implementation in affordable hardware devices like microcontrollers. Finally, the proposed method result is compared with U-Net, DeepLabV3+ with MobileNet-V2 and DeepLabV3+ with ResNet-50. [ABSTRACT FROM AUTHOR]

Sensors have different perceptive abilities against environment. Sensor fusion plays a crucial role at achieving better perception by accumulating the information acquired at different times. But, the decision of the observation may conflict with each other due to usage of different algorithms, thresholds on processing algorithms, and different perceptive character of sensors. This study presents the late fusion method applied to outputs provided by deep learning models fed by camera and lidar sensor's measurement data. For camera sensor, a deep learning model as a multi-task network is proposed to multi-classify cars, motorcycles, bicycles, buses, trucks, and pedestrians under the category of dynamic traffic objects. In addition, color classified traffic lights and traffic signs with a capability of segmenting drivable area and detecting lane lines are classified under the category of static traffic objects. The proposed multi-network is trained and tested with the BDD100K dataset and benchmarked with publicly available multi-networks. The presented method is the second fastest multi-network reaching at 52 FPS runtime, ranked second based on the drivable area segmentation and lane line detection performance. For segmentation of dynamic objects, the network performance is increased by 22.45%, and considering mIoU overall performance increase is 3.96%. For a lidar sensor, a different modality is presented to detect objects. Two sensors' data are fused by proposed fusion algorithm, and results are tested and evaluated with the KITTI dataset. The proposed fusion methodology outperforms the stand-alone lidar methods about 3.58% and 3.63% on BEV and 3D detection MAP, respectively. Overall, benchmarking with two distinct fusion approaches illustrates the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Long time series forecasting (LTSF), which involves modeling relationships within long time series to predict future values, has extensive applications in domains such as weather forecasting, financial analysis, and traffic prediction. Recently, numerous transformer-based models have been developed to address the challenges in LTSF. These models employ methods such as sparse attention to alleviate the inefficiencies associated with the attention mechanism and utilize decomposition architecture to enhance the predictability of the series. However, these complexity reduction methods necessitate additional calculations, and the series decomposition architecture overlooks the random components. To overcome these limitations, this paper proposes the Frequency Enhanced Decomposed Attention Free Transformer (FEDAF). FEDAF introduces two variants of the Frequency Enhanced Attention Free Mechanism (FEAFM), namely FEAFM-s and FEAFM-c, which seamlessly replace self-attention and cross-attention. Both variants perform calculations in the frequency domain without incurring additional costs, with the time and space complexity of FEAFM-s being O (L log L) . Additionally, FEDAF incorporates a time series decomposition architecture that considers random components. Unlike other models that solely decompose the series into trend and seasonal components, FEDAF also eliminates random terms by applying Fourier denoising. Our study quantifies data drift and validates that the proposed decomposition structure can mitigate the adverse effects caused by data shift. Overall, FEDAF demonstrates superior forecasting performance compared to state-of-the-art models across various domains, achieving a remarkable improvement of 19.49% for Traffic in particular. Furthermore, an efficiency analysis reveals that FEAFM enhances space efficiency by 12.8% compared to the vanilla attention mechanism and improves time efficiency by 43.63% compared to other attention mechanism variants. [ABSTRACT FROM AUTHOR]

This study addresses the importance of enhancing traditional fluid-flow solvers by introducing a Machine Learning procedure to model pressure fields computed by standard fluid-flow solvers. The conventional approach involves enforcing pressure–velocity coupling through a Poisson equation, combining the Navier–Stokes and continuity equations. The solution to this Poisson equation constitutes a substantial percentage of the overall computational cost in fluid flow simulations, therefore improving its efficiency can yield significant gains in computational speed. The study aims to create a versatile method applicable to any geometry, ultimately providing a more efficient alternative to the conventional pressure solver. Machine Learning models were trained with flow fields generated by a Computational Fluid Dynamics solver applied to the confined flow over multiple geometries, namely wall-bounded cylinders with circular, rectangular, triangular, and plate cross-sections. To achieve applicability to any geometry, a method was developed to estimate pressure fields in fixed-shape blocks sampled from the flow domain and subsequently assemble them to reconstruct the entire physical domain. The model relies on multilayer perceptron neural networks combined with Principal Component Analysis transformations. The developed Machine Learning models achieved acceptable accuracy with errors of around 3%. Furthermore, the model demonstrated enhanced computational efficiency, outperforming the classical PISO algorithm by up to 30 times. [ABSTRACT FROM AUTHOR]

Due to the uncertainty of geological conditions during the tunneling process, advanced prediction of TBM tunneling parameters is significant for evaluating operational safety and efficiency, especially for real-time prediction of key tunneling parameters during the steady phase of TBM operation. At present, although there are studies on constructing predictive models based on machine learning algorithms, multiparameter prediction consistent with the actual tunneling process remains challenging due to the complexity of the TBM tunneling process and the numerous tunneling parameters. Therefore, this paper proposes a real-time multiple tunneling parameters prediction method of TBM steady phase based on dual recurrent neural networks. Firstly, the irregular multidimensional time series of tunneling parameters are analyzed and processed, which are divided into an idle-push phase, a rising phase, and a steady phase; secondly, the parameters of rising phase are analyzed using a recurrent neural network, and the parameters relevant for constructing a real-time prediction model are screened; then, based on the screened parameters, the Bayesian-optimized gated recurrent unit (GRU, a kind of recursive neural network) is proposed to construct a real-time prediction model for the four key tunneling parameters during the steady phase. Finally, the effectiveness and practicality of the proposed method are demonstrated by verification on real TBM tunnel datasets and comparing it with the models constructed by six commonly used machine learning algorithms. The results of this paper show that the designed prediction method is able to achieve a good combination of performance in terms of accuracy and computational time-consumption, with an average prediction accuracy of 91.1% for the four parameters for different rock grades of geology, the multiparameter prediction time for 100 samples is only 11 ms. In addition, three current similar studies using deep learning methods were compared to demonstrate the superiority of this proposal. As a method more closer to practical application, this work provides guidance for the forward-looking prediction of TBM tunneling parameters. [ABSTRACT FROM AUTHOR]

In light of current developments in dental care, dental professionals have increasingly used deep learning methods to get precise diagnoses of oral problems. Using intraoral X-rays in dental radiography is imperative in many dental interventions. Integrating deep learning techniques with a unique collection of intraoral X-ray images has been undertaken to enhance the accuracy of dental disease detection. In this study, we propose an alternative pooling layer, namely the Common Vector Approach Pooling technique, to address the constraints associated with average pooling in deep learning methods. The experiments are conducted on a large dataset, involving twenty different dental conditions, divided into seven categories. Our proposed approach achieved a high accuracy rate of 86.4% in identifying dental problems across the seven oral categories. [ABSTRACT FROM AUTHOR]

Soil volumetric water content (VWC) is one of the key factors in hydrological cycles and responsible for inducing droughts and floods. Therefore, the precise prediction of VWC is crucial for the effective management of water resources. However, the complexity in structural characteristics and interaction with several other external meteorological factors cause difficulty in establishing a mathematical model which can predict soil VWC accurately. This study demonstrates the applicability of Convolution Neural Network-Long short-term memory (CNN-LSTM) hybrid model to predict soil VWC (%), concentrating specifically on optimizing the predictors combination using recursive feature elimination (RFE) which results in a more interpretable model with less complexity. The model was developed using the data collected from Benton County of Washington, USA, and generalization capacity of the model was tested in other counties of Washington. To verify the improved prediction ability of the proposed model, the results were compared with the established CNN, LSTM and MLR models. The results reflected that the proposed CNN-LSTM model predicted better than the individual CNN, LSTM and MLR models for the training site as well as for the five testing sites, proving its good generalization capacity. [ABSTRACT FROM AUTHOR]

Retinal fundus image classification is pivotal for the early detection of various eye disorders. Leveraging advancements in deep learning, this research introduces a robust framework for retinal fundus image classification. In this research paper, we introduce a novel methodology that integrates the Active Gradient Deep Convolutional Neural Network (AG-DCNN) architecture with the Red Spider Optimization (RSO) algorithm. Extensive evaluations conducted on well-established retinal image datasets such as DRIVE, STARE, CHASE DB1, HRF, DRISHTI-GS, and RFMiD illustrate substantial enhancements in classification accuracy, sensitivity, and specificity when compared to conventional approaches. The AG-DCNN with RSO exhibits superior performance and remarkable generalization abilities across diverse datasets. This novel approach not only enhances the latest advances in retinal image classification but also holds promise for early disease diagnosis and improved patient care. Through extensive examinations, our proposed method consistently outperforms established techniques, establishing itself as the benchmark in retinal fundus image classification. The integration of AG-DCNN and RSO showcases its efficacy and potential impact in advancing the field of medical image analysis. [ABSTRACT FROM AUTHOR]

Deep learning applications have witnessed significant advancements across diverse domains, revolutionizing tasks such as image recognition, disease classification, cancer detection, natural language processing, and autonomous decision-making. However, striking a balance between performance and cost remains a fundamental challenge. Performance, which encompasses accuracy, efficiency, and effectiveness, is crucial for the real-world applicability of deep learning models. Attaining high performance often involves using resource-intensive models with intricate architectures, extensive datasets, and hyperparameter tuning. While these factors lead to improved performance, they also result in increased costs, including computational resources, time, and energy consumption. This research addresses this challenge by proposing a novel low-cost deep learning model based on Sharpened Cosine Similarity (LC-SCS) for breast cancer classification using thermal images. In our study on a DMR-IR dataset, we also utilized pre-trained models: ResNet-101, VGG-16, Inception-V3, ResNet-50, VGG-19, and Xception. The proposed LC-SCS model achieved an impressive accuracy of 94%, trailing just 4% behind the leading VGG-19 model, while maintaining a low computational cost. It achieved recall and precision scores of 0.95 each, with an F1-score of 0.94. The LC-SCS model excels in computational efficiency, demanding only 1.85 GFLOPs and 1.2 million parameters. The memory requirement is minimal at 865.35 MB, and it exhibits a latency of just 0.003 s. Additionally, the CPU execution time for prediction is 3.59 s. Comparatively, the best-performing pre-trained VGG-19 model achieved 98% accuracy but incurred significantly higher costs. The proposed LC-SCS model showcases a remarkable 298.63 times reduction in GFLOPs, a 116.31 times reduction in total parameters, a 4.57 times reduction in memory requirements for model weights, a 5 times reduction in latency, and a 3.33 times reduction in CPU execution time for prediction, thus making it as an exceptionally resource-efficient model, particularly in scenarios with limited computational resources. [ABSTRACT FROM AUTHOR]

Automatic skin lesion subtyping is a crucial step for diagnosing and treating skin cancer and acts as a first level diagnostic aid for medical experts. Although, in general, deep learning is very effective in image processing tasks, there are notable areas of the processing pipeline in the dermoscopic image regime that can benefit from refinement. Our work identifies two such areas for improvement. First, most benchmark dermoscopic datasets for skin cancers and lesions are highly imbalanced due to the relative rarity and commonality in the occurrence of specific lesion types. Deep learning methods tend to exhibit biased performance in favor of the majority classes with such datasets, leading to poor generalization. Second, dermoscopic images can be associated with irrelevant information in the form of skin color, hair, veins, etc.; hence, limiting the information available to a neural network by retaining only relevant portions of an input image has been successful in prompting the network towards learning task-relevant features and thereby improving its performance. Hence, this research work augments the skin lesion characterization pipeline in the following ways. First, it balances the dataset to overcome sample size biases. Two balancing methods, synthetic minority oversampling TEchnique (SMOTE) and Reweighting, are applied, compared, and analyzed. Second, a lesion segmentation stage is introduced before classification, in addition to a preprocessing stage, to retain only the region of interest. A baseline segmentation approach based on Bi-Directional ConvLSTM U-Net is improved using conditional adversarial training for enhanced segmentation performance. Finally, the classification stage is implemented using EfficientNets, where the B2 variant is used to benchmark and choose between the balancing and segmentation techniques, and the architecture is then scaled through to B7 to analyze the performance boost in lesion classification. From these experiments, we find that the pipeline that balances using SMOTE and uses the adversarially trained segmentation network achieves the best baseline performance of 91% classification accuracy with EfficientNet B2. Based on the scaling experiments, we find that optimal performance is reached with the B6 architecture that classifies with a 97% accuracy. Furthermore, the proposed pipeline for lesion characterization outperforms the state of the art performance on the ISIC dataset. [ABSTRACT FROM AUTHOR]

Efficient flood risk management hinges on the precise mapping and assessment of areas vulnerable to flooding. This research endeavors to advance the flood susceptibility mapping in Jeddah, Saudi Arabia by harnessing the long short-term memory (LSTM) algorithm enriched with two sophisticated metaheuristic optimizers: invasive weed optimization (IWO) and harmony search (HS). The process commenced with the utilization of synthetic aperture radar (SAR) imagery to construct a detailed flood inventory map. A comprehensive geodatabase encompassing various flood conditioning factors—encompassing lithology, land use, proximity to water bodies, hydrologic soil group (HSG), topographical features (such as slope, plan curvature, and aspect), and hydrological indices [including profile curvature, Topographic Wetness Index (TWI), flow accumulation, Topographic Position Index (TPI), altitude, Terrain Ruggedness Index (TRI), and Stream Power Index (SPI)] was meticulously curated. To develop the model, 70% of this dataset was employed, while the remaining 30% served to validate the predictive efficacy of the resultant flood susceptibility maps. These maps' accuracy was quantitatively gauged through the receiver operating characteristic curve and the area under the curve (AUC) statistics. Findings reveal that the integration of LSTM with IWO and HS metaheuristic algorithms significantly enhances accuracy (LSTM-IWO: AUC = 0.900; LSTM-HS: AUC = 0.876) in comparison to the standalone LSTM approach (AUC = 0.863). The implementation of these hybrid algorithms manifests as a potent and economically viable approach for detailed geospatial modeling of flood susceptibility, providing invaluable insights to bolster flood mitigation, preparedness, and emergency response strategies. [ABSTRACT FROM AUTHOR]

Most operational weather and climate models carry out forecasts or simulations of atmospheric variables at low spatial resolutions, but we often need high-resolution projections of such variables. The process of mapping low-resolution projections to high-resolution projections is called spatial downscaling. This is analogous to the computer vision task of single-image super-resolution (SISR). In recent studies, convolution-based architectures including UNet and its variants have emerged as a good choice for SISR. Since the gridded spatial map of any climate variable is analogous to an image, we can use the SISR-based models for spatial downscaling. In this paper, we present a novel UNet-based architecture called max-average UNet (MAUNet) to downscale the precipitation. We have proposed Max-Average Units (MAUs) that include a max-pooling, an average-pooling, and an averaging unit for the encoding part of the UNet. For the decoding part, we develop UpSampler Unit (USU), which too utilizes averaging. We demonstrate the importance of max-averaging units through toy experiments on the MNIST and CelebA dataset. We also examine the role of dropouts for this task and experimentally demonstrate their skill to improve the model's performance on different datasets. The task examined here is to obtain fourfold resolution enhancement of monsoon precipitation over the Indian landmass and also over Southeast Continental USA (CONUS) region. The models are evaluated using RMSE, PSNR, MSSIM, and correlation coefficient as evaluation matrices. We have made detailed comparisons to show that MAUNet can produce superior downscaling than standard interpolation techniques as well as previously used deep learning models like UNet, EDSR, and SRDRN. [ABSTRACT FROM AUTHOR]

In an increasingly complex and volatile environment, government officials, researchers, and investors alike would like to possess models that accurately forecast markets in order to make appropriate decisions. This research investigates the efficacy of Transformer-based deep neural networks in predicting financial market returns compared to traditional models, focusing on ten different market indexes. The study employs a comprehensive methodology that involves iterative dropout tests and batch size optimization to enhance model performance. By leveraging the power of deep learning, the research aims to improve prediction accuracy and capture complex patterns in stock market data. Twelve neural network architectures are compared across ten indexes to measure their performance, finding that the proposed Transformer variants produce significantly better results compared to benchmark models in all cases. The results of ablative experiments reveal the superiority of Transformer models in capturing long-term dependencies and extracting meaningful features from time series data. The findings suggest that Transformer neural networks outperform LSTM networks and other traditional models in forecasting financial market trends. This research contributes to the growing body of literature on deep learning applications in finance and provides valuable insights for government officials, researchers, and investors seeking to make informed decisions in the stock market. The implications of this study extend beyond academia, offering practical implications for enhancing prediction accuracy and optimizing investment strategies in the dynamic and volatile financial market landscape. [ABSTRACT FROM AUTHOR]

Stock price prediction is a significant research domain, intersecting statistics, finance, and economics. Accurately forecasting stock price trends has always been a focal point for many researchers. However, traditional statistical methods for time series prediction still lack accuracy. The existing deep learning-based methods for stock price prediction have significantly enhanced the accuracy of predicting individual stock prices. However, they are not effective in forecasting the probability range of future stock price trends. In this paper, to address these limitations, we propose a novel DeepAR model based on the attention mechanism (DeepARA) for both single-point and probabilistic predictions of stock prices. This enhances the accuracy and flexibility of stock price forecasting. Although the attention mechanism was initially developed for natural language processing, it has now found applications in time series forecasting, including the dynamics of the stock market. Attention allocates different weights to time points of varying importance, thereby enhancing the model's ability to capture fundamental market dynamics. We conducted multiple experiments in the Chinese stock market, involving 30 stocks across the top six sectors. Compared with baseline models, the DeepARA model demonstrates superior predictive capabilities. [ABSTRACT FROM AUTHOR]

This paper introduces a new learning-based approach for approximately solving the Kidney-Exchange Problem (KEP), an NP-hard problem on graphs. The KEP consists of, given a pool of kidney donors and patients waiting for kidney donations, optimally selecting a set of donations to optimize the quantity and quality of transplants performed while respecting a set of constraints about the arrangement of these donations. The proposed technique consists of two major steps: the first is a Graph Neural Network (GNN) trained without supervision; the second is a deterministic non-learned search heuristic that uses the output of the GNN to find a valid solution. To allow for comparisons, we also implemented and tested an exact solution method using integer programming, two greedy search heuristics without the machine learning module, and the GNN alone without a heuristic. We analyze and compare the methods and conclude that the learning-based two-stage approach is the best solution quality, outputting approximate solutions on average 1.1 times more valuable than the ones from the deterministic heuristic alone. [ABSTRACT FROM AUTHOR]

One of the critical challenges in brain-computer interfaces is the classification of brain activities through the analysis of EEG signals. This paper seeks to improve the efficacy of deep learning-based rehabilitation systems, aiming to deliver superior services for individuals with physical disabilities. The research introduces the time distributed long short-term memory (TD-LSTM) framework, which incorporates an LSTM and a time distributed approach to classify brain activities. Learning in TD-LSTM is achieved by uncovering time-dependent semantic dependencies within EEG signals over time. By extracting all discriminative and relevant spatiotemporal dependencies via TD-LSTM, valuable information on different time steps in each sequence has been obtained. Time distributed approach shortens the input time series, making learning from long time series sequences easier, and the learning process of complex temporal and spatial dependencies in time series multi-channel EEG signals becomes more efficient. The main contributions in this paper can be outlined as follows: (1) implementation of brain activity binary classification of motor imagery/execution tasks using time distributed approach via RNN network for the first time, (2) evaluation of the performance and generalizability of the proposed method on four benchmark datasets, (3) dual-purpose classification which providing an efficient ways for classifying both types of motor imagery/execution brain activity. The experimental results show that the proposed method performs well compared to several baseline research works. [ABSTRACT FROM AUTHOR]

Tuberculosis caused by the infection of Mycobacterium. It is the fifth major source of death and one of the greatest threats to humans in the modern world. Thus, it needs to be detected at an earlier stage using the chest X-rays (CXR) image for precise identification and treatment. The suggested scheme's main goal is to identify this deadly disease using CXR with improved classification accuracy. This detection process comprises pre-processing, noise removal, balancing of image level, application of the Double Attention Res-U-Net-based Deep Neural Network (DARUNDNN) model, and optimization of deep learning features using the Dingo Optimization Algorithm for achieving better accuracy. The experimental validation of the proposed DARUNDNN model is conducted using benchmark datasets, namely Montgomery, Shenzhen, and National Institutes of Health CXR images. The results obtained using the Shenzhen dataset confirm that the proposed DARUNDNN model is efficient in achieving better accuracy of 98.92%, specificity of 97.24%, and sensitivity of 98.86% with a least error of 1.6 compared to the benchmarked models used for investigation. Moreover, the experimental validation conducted using the Montgomery County dataset also confirmed an excellent accuracy of 98.982%, a specificity of 97.56%, and a sensitivity of 98.52%, with a least error of 1.32 compared to the baseline approaches used for investigation. [ABSTRACT FROM AUTHOR]

The survival rate of lung cancer relies significantly on how far the disease has spread when it is detected, how it reacts to the treatment, the patient's overall health, and other factors. Therefore, the earlier the lung cancer diagnosis, the higher the survival rate. For radiologists, recognizing malignant lung nodules from computed tomography (CT) scans is a challenging and time-consuming process. As a result, computer-aided diagnosis (CAD) systems have been suggested to alleviate these burdens. Deep-learning approaches have demonstrated remarkable results in recent years, surpassing traditional methods in different fields. Researchers are currently experimenting with several deep-learning strategies to increase the effectiveness of CAD systems in lung cancer detection with CT. This work proposes a deep-learning framework for detecting and diagnosing lung cancer. The proposed framework used recent deep-learning techniques in all its layers. The autoencoder technique structure is tuned and used in the preprocessing stage to denoise and reconstruct the medical lung cancer dataset. Besides, it depends on the transfer learning pre-trained models to make multi-classification among different lung cancer cases such as benign, adenocarcinoma, and squamous cell carcinoma. The proposed model provides high performance while recognizing and differentiating between two types of datasets, including biopsy and CT scans. The Cancer Imaging Archive and Kaggle datasets are utilized to train and test the proposed model. The empirical results show that the proposed framework performs well according to various performance metrics. According to accuracy, precision, recall, F1-score, and AUC metrics, it achieves 99.60, 99.61, 99.62, 99.70, and 99.75%, respectively. Also, it depicts 0.0028, 0.0026, and 0.0507 in mean absolute error, mean squared error, and root mean square error metrics. Furthermore, it helps physicians effectively diagnose lung cancer in its early stages and allows specialists to improve the accuracy and consistency of workflow. [ABSTRACT FROM AUTHOR]

Due to the heterogeneity gap in multi-view data, researchers have been attempting to apply these data to learn a co-latent representation to bridge this gap. However, multi-view representation learning still confronts two challenges: (1) it is hard to simultaneously consider the performance of downstream tasks and the interpretability and transparency of the network; (2) it fails to learn representations that accurately describe the class boundaries of downstream tasks. To overcome these limitations, we propose an interpretable representation learning framework, named interpretable multi-view proximity representation learning network. On the one hand, the proposed network is customized by an explicitly designed optimization objective that enables it to learn semantic co-latent representations while maintaining the interpretability and transparency of the network from the design level. On the other hand, the designed multi-view proximity representation learning objective function encourages its learned co-latent representations to form intuitive class boundaries by increasing the inter-class distance and decreasing the intra-class distance. Driven by a flexible downstream task loss, the learned co-latent representation can adapt to various multi-view scenarios and has been shown to be effective in experiments. As a result, this work provides a feasible solution to a generalized multi-view representation learning framework and is expected to accelerate the research and exploration in this field. [ABSTRACT FROM AUTHOR]

In the federated learning scenario, the private data are kept local, and gradients are shared to train the global model. Because gradients are updated according to the private training data, the features of the data are encoded into gradients. Prior work proved the possibility of reconstructing the private training data based on gradients. However, only a small batch of images can be recovered, and the reconstruction quality, especially against the large batch size of images, is unsatisfactory. To improve the quality of reconstruction of a large batch of images, a generative gradient inversion attack based on a regulation term is designed, which is called fDLG. First, a regulation term that can avoid drastic variations within image regions is proposed, which is based on the cognition that changes between image pixels are gradual. The proposed regulation term encourages the synthesized dummy image to be piece-wise smooth. Second, generative adversarial networks are trained to improve the quality of the attack with the global model used as a discriminator. Simulation shows that large batches of images (128 images on CIFAR100, 256 images on MNIST) can be faithfully reconstructed at high resolution, and even large images from ImageNet can be reconstructed. [ABSTRACT FROM AUTHOR]

In cybersecurity, live production data for predictive analysis pose a significant challenge due to the inherently secure nature of the domain. Although there are publicly available, synthesized, and artificially generated datasets, authentic scenarios are rarely encountered. For anomaly-based detection, the dynamic definition of thresholds has gained importance and attention in detecting abnormalities and preventing malicious activities. Unlike conventional threshold-based methods, deep learning data modeling provides a more nuanced perspective on network monitoring. This enables security systems to continually refine and adapt to the evolving situation in streaming data online, which is also our goal. Furthermore, our work in this paper contributes significantly to AIOps research, particularly through the deployment of our intelligent module that cooperates within a monitoring system in production. Our work addresses a crucial gap in the security research landscape toward more practical and effective secure strategies. [ABSTRACT FROM AUTHOR]

The integration of cloud and edge computing, along with machine learning, plays a vital role in the development of efficient healthcare systems in smart cities. However, machine and deep learning (DL) models are prone to delayed convergence and Type-I and Type-II errors due to data vastness and high degree imbalance. To overcome the shortcomings of previous frameworks, this work aims to propose an optimization method with DL, 'Deep-Chaotic Nature Whale Optimization' (Deep-CNWO) for early prediction of Blood Pressure disorders among patients under at-home supervision. A simplex search algorithm is integrated to improve the update mechanism of whale optimization algorithm (WOA), thereby creating a CNWO algorithm. The purpose of this hybrid optimization is to increase the accuracy and efficiency of DL models. Leveraging the power of DL and CNWO, this method (Deep-CNWO) provides an effective solution for early detection and proactive management of a chronic disease in at-home healthcare settings. We collected relevant data from clinical studies, including vital signs and patient contextual information, to train and evaluate the deep-CNWO model. The CNWO optimization approach has been used to improve the predictive performance and convergence of DL models. Experiments performed on imbalanced datasets using deep-CNWO have given 99.90% accuracy. The average F-score for emergency cases has improved by 22%, while the average accuracy has increased by 5.72% across all three classes, compared to the results reported in previous related work. Deep-CNWO improves the convergence of DL and reduces Type-I and Type-II errors. The experimental results demonstrate the efficacy of our proposed method for remote patient monitoring and highlight its potential for quick intervention during emergencies. [ABSTRACT FROM AUTHOR]

Deception detection is an interdisciplinary field attracting researchers from psychology, criminology, computer science, and economics. Automated deception detection presents unique challenges compared to traditional polygraph tests, but also offers novel economic applications. In this spirit, we propose an approach combining deep learning with discriminative models for deception detection. Therefore, we train CNNs for the facial modalities of gaze, head pose, and facial expressions, allowing us to compute facial cues. Due to the very limited availability of training data for deception, we utilize early fusion on the CNN outputs to perform deception classification. We evaluate our approach on five datasets, including four well-known publicly available datasets and a new economically motivated rolling dice experiment. Results reveal performance differences among modalities, with facial expressions outperforming gaze and head pose overall. Combining multiple modalities and feature selection consistently enhances detection performance. The observed variations in expressed features across datasets with different contexts affirm the importance of scenario-specific training data for effective deception detection, further indicating the influence of context on deceptive behavior. Cross-dataset experiments reinforce these findings. Notably, low-stake datasets, including the rolling dice Experiment, present more challenges for deception detection compared to the high-stake Real-Life trials dataset. Nevertheless, various evaluation measures show deception detection performance surpassing chance levels. Our proposed approach and comprehensive evaluation highlight the challenges and potential of automating deception detection from facial cues, offering promise for future research. [ABSTRACT FROM AUTHOR]

The thermal efficiency for the pelletizing process is intrinsically linked to the diameter and humidity of the iron ore pellets, so that the sensing of the granulometric range in the formation of the pellets becomes essential to the flow of the pelletizing process in the steel industry; this paper presents the assembly of a computer vision system for the detection and segmentation of pellets aiming at the automation of the granulometric measurement, following its formation in the pelletizing disk, using the instance segmentation method to verify whether the particle granulometric distribution (PSD) is adequate for "real-time" applications. The system calculates the normal distribution of the diameter in millimeters, evaluating the normal curve and the standard deviation of the segmented pellets, using a deep neural network based on the You Only Look At CoefficienTs (YOLACT) network, adding speed and precision in the granulometric analysis. In the sample sets, the need for adjustment factors inherent to the pelletizing process became evident. This led to the establishment of the computer vision system, termed the Volumetric Correction Factor (VCF) and Visual Overlay Factor (VOF). The VCF is utilized to estimate the volume of pellets within the pelletizing disk during operation, while the VOF adjusts the millimeter-per-pixel (mpp) ratio. The results of the measurement system proved to be efficient in real-time granulometric measurement. [ABSTRACT FROM AUTHOR]

Ultrasound is a readily available, non-invasive and low-cost screening for the identification of endometriosis lesions, but its diagnostic specificity strongly depends on the experience of the operator. For this reason, computer-aided diagnosis tools based on Artificial Intelligence techniques can provide significant help to the clinical staff, both in terms of workload reduction and in increasing the overall accuracy of this type of examination and its outcome. However, although these techniques are spreading rapidly in a variety of domains, their application to endometriosis is still very limited. To fill this gap, we propose and evaluate a novel multi-scale ensemble approach for the automatic segmentation of endometriosis lesions from transvaginal ultrasounds. The peculiarity of the method lies in its high discrimination capability, obtained by combining, in a fusion fashion, multiple Convolutional Neural Networks trained on data at different granularity. The experimental validation carried out shows that: (i) the proposed method allows to significantly improve the performance of the individual neural networks, even in the presence of a limited training set; (ii) with a Dice coefficient of 82%, it represents a valid solution to increase the diagnostic efficacy of the ultrasound examination against such a pathology. [ABSTRACT FROM AUTHOR]

Visual question answering is a challenging task that necessitates for sophisticated reasoning over the visual elements to provide an accurate answer to a question. Majority of the state-of-the-art VQA models are only applicable to English questions. However, applications such as visual assistance and tourism necessitate the incorporation of multilingual VQA systems. This paper presents an effective deep learning framework for Malayalam visual question answering (MVQA), which can answer a specific natural language question about an image in Malayalam. As there is no available dataset in English–Malayalam VQA, a MVQA dataset was created by translating English question–answer pairs from the visual genome dataset. The paper proposes an attention-driven MVQA model on the developed dataset. The proposed MVQA model uses a deep learning-based co-attention mechanism to jointly learn the attention for images and Malayalam questions. A second-order multimodal factorized high-order pooling is used for multi modal feature fusion. Different VQA models using combinations of classical CNNs and RNNs were experimented on the developed MVQA dataset, and the performance was compared against the proposed attention-driven model. Experimental results show that the proposed attention-driven MVQA model achieves state-of-the-art results as compared to other models for MVQA on the custom Malayalam VQA dataset. [ABSTRACT FROM AUTHOR]

Breast cancer poses a threat to women's health and contributes to an increase in mortality rates. Mammography has proven to be an effective tool for the early detection of breast cancer. However, it faces many challenges in early breast cancer detection due to poor image quality, traditional segmentation, and feature extraction. Therefore, this work addresses these issues and proposes an attention-based backpropagation convolutional neural network (ABB-CNN) to detect breast cancer from mammogram images more accurately. The proposed work includes image enhancement, reinforcement learning-based semantic segmentation (RLSS), and multiview feature extraction and classification. The image enhancement is performed by removing noise and artefacts through a hybrid filter (HF), image scaling through a pixel-based bilinear interpolation (PBI), and contrast enhancement through an election-based optimization (EO) algorithm. In addition, the RLSS introduces intelligent segmentation by utilizing a deep Q network (DQN) to segment the region of interest (ROI) strategically. Moreover, the proposed ABB-CNN facilitates multiview feature extraction from the segmented region to classify the mammograms into normal, malignant, and benign classes. The proposed framework is evaluated on the collected and the digital database for screening mammography (DDSM) datasets. The proposed framework provides better outcomes in terms of accuracy, sensitivity, specificity, precision, f-measure, false-negative rate (FNR) and area under the curve (AUC). This work achieved (99.20%, 99.35%), (99.56%, 99.66%), (98.96%, 98.99%), (99.05%, 99.12%), (0.44%, 0.34%), (99.31%, 99.39%) and (99.27%, 99.32%) of accuracy, sensitivity, specificity, precision, FNR, f-measure and AUC on (collected, DDSM datasets), respectively. This research addresses the prevalent challenges in breast cancer identification and offers a robust and highly accurate solution by integrating advanced deep-learning techniques. The evaluated results reveal the proposed framework's potential in early breast cancer detection. [ABSTRACT FROM AUTHOR]

With the wide application of deep learning in anomaly detection (AD), industrial vision AD has achieved remarkable success. However, current AD usually focuses on anomaly localization and rarely investigates anomaly classification. Furthermore, anomaly classification is currently requested for quality management and anomaly reason analysis. Therefore, it is essential to classify anomalies while improving the accuracy of AD. This paper designs a novel multi-classification AD (MCAD) framework to achieve high-accuracy AD with an anomaly classification function. In detail, the proposal model based on relational knowledge distillation consists of two components. The first one employs a teacher–student AD model, utilizing a relational knowledge distillation approach to transfer the interrelationships of images. The teacher–student critical layer feature activation values are used in the knowledge transfer process to achieve anomaly detection. The second component realizes anomaly multi-classification using the lightweight convolutional neural network. Our proposal has achieved 98.95, 96.04, and 92.94% AUROC AD results on MNIST, FashionMNIST, and CIFAR10 datasets. Meanwhile, we earn 97.58 and 98.10% AUROC for AD and localization in the MVTecAD dataset. The average classification accuracy of anomaly classification has reached 76.37% in fifteen categories of the MVTec-AD dataset. In particular, the classification accuracy of the leather category has gained 95.24%. The results on the MVTec-AD dataset show that MCAD achieves excellent detection, localization, and classification results. [ABSTRACT FROM AUTHOR]

Spices and other food products have been permanently susceptible to adulteration, affecting safety and acceptability when commercialized. A relevant alternative to detect contaminants in food products is to couple near-infrared spectroscopy (NIR) with chemometrics. Among the most accurate chemometric techniques employed to analyze food products, partial least squares regression (PLSR) combines features from and generalizes principal component analysis (PCA) to create compact and accurate models. Other techniques inspired in the human brain, such as multilayer perceptron, the long short-term memory (LSTM) models, and other approaches based on deep learning, take advantage of the high complexity of weights and neurons to train models based on large amounts of data. In this paper, a methodology is proposed to evaluate chemometric tools to estimate the percentage of adulterants in paprika powder using NIR spectroscopy, and three approaches are proposed and compared showing different performances. According to the methodology, the paprika samples were dried and separated into pericarp, peduncle, and seed cake. The resulting elements were finely milled, sieved, and mixed into 21 different combinations with a different percentage of each. Spectral profiles were used to train PLSR, multilayer perceptron, and regression models based on LSTM networks. The models were compared following a k-fold cross-validation strategy. Results showed that PLSR presented the highest R 2 = 0.978 for peduncle adulterant estimation, and the lowest R M S E = 6.24 . In particular, when seed cake powder was used as an adulterant, the PLSR approach showed the highest R 2 = 0.981 , and the lowest R M S E = 5.806 . The RPD values were higher than 2.000 for all models that use the peduncle as an adulterant and only for models bound to the PLSR in the adulterated samples with pressed seed cake. In summary, the best predictions were obtained using PLSR models, providing evidence of the feasibility of using NIR spectra to estimate the percentage of adulterants in paprika powder. [ABSTRACT FROM AUTHOR]

Adopting an efficient software process model is critical for building high-quality software applications. An important factor impacting the software development process is an accurate estimate of human effort required to complete the software project. While machine learning methods were historically used to develop estimation models, there has been little investigation into the potential of deep convolutional neural networks (DCNNs) for improving software effort estimation. One of the biggest obstacles in using DCNN for this purpose is the common nature of software datasets, which often consist of vectorized samples rather than matrices. To defeat this obstacle and reduce vagueness in software attribute measurement, this study uses Fuzzy theory to generate an appropriate two-dimensional datapoint representation. The fuzzy clustering is commonly used to split dataset samples into separate clusters, which can help to generate Fuzzy membership functions. This approach makes it easier to generate a two-dimensional array representation for each data sample based on the membership values, allowing it to be used as input to the DCNN model. The efficiency of the proposed model was thoroughly evaluated using PROMISE benchmark datasets. The findings based on mean absolute errors and standardized accuracy show that our proposed model produced very good performance with low error rates and outperformed several current state-of-the-art effort estimation models. Nonetheless, further research is needed to determine the impact of different cluster numbers and features on the performance of our model. In conclusion, this study emphasizes the possibility for incorporating DCNN into software effort estimates and highlights the viability of utilizing fuzzy modeling and clustering techniques to enhance the data representation of software datasets. [ABSTRACT FROM AUTHOR]

Pathologic myopia (PM) is the critical factor of irreversible visual artifacts and puts patients at risk of other severe retinal diseases such as glaucoma. Early intervention can help control the disease's progression and prevent vision loss. Due to its prevalence worldwide, automated detection of PM and its severity is essential. Deep learning-aided diagnosis has proven itself in the field of ophthalmology. The proposed study automatically classifies pathologic and non-pathologic myopia from the fundus images using a guided mini U-Net (GM-U-Net) for feature extraction integrated with a customized convolutional neural network (PMNet) explicitly designed for fundus images. The proposed GM-U-Net allows a deeper network with significantly reduced parameters than conventional U-Net for feature extraction. The proposed PMNet utilizes the features extracted by GM-U-Net to discriminate between PM and a normal retina image. The PMNet classification performance is compared with the other transfer learning models based on the features provided by the GM-U-Net. The combination of GM-U-Net and PMNet outperforms the different models for PM classification. In-depth ablation tests are conducted to realize the current form of PMNet and test its effectiveness. PMNet achieves an accuracy of 90%, average sensitivity of 93%, and specificity of 97% for binary class on the test set, demonstrating it as a valuable tool for early PM detection. Further, to localize the prominent regions in the images, colored heatmap techniques are applied to visualize the affected areas with a hotter color. [ABSTRACT FROM AUTHOR]

The oil and gas industry employs numerical simulation tools extensively in reservoir analysis and strategic planning. This study presents a machine-learning proxy model, employing a Few-shot Learning approach with a Deep Convolutional Generative Adversarial Network (DC-GAN) to reduce computational costs using fewer training samples. The DC-GAN generates new training samples by synthesizing spatial attributes, reservoir parameters, and time series data, thus enhancing the sample variability and diversity needed for accurate production prediction. Also, the study proposes a straightforward and efficient method for data augmentation that primarily involves replicating the initial training dataset. The accumulated production forecast generated from geostatistical realizations enables intelligent reservoir management through risk analysis. The technique can reduce the processing footprint by 70%. Unlike most reservoir studies that employ synthetic datasets, this investigation adopts a real, high-dimensional, and complex reservoir model. [ABSTRACT FROM AUTHOR]

Retinal diseases that are not treated in time can cause irreversible, permanent damage, including blindness. Although a patient may suffer from more than one retinal disease at the same time, most of the studies focus on the diagnosis of a single disease only. Therefore, to detect multi-label retinal diseases from color fundus images, we developed an end-to-end deep learning architecture that combines the EfficientNet backbone with the ML-Decoder classification head in this study. While EfficientNet provides powerful feature extraction with fewer parameters via compound scaling, ML-Decoder further improves efficiency and flexibility by reducing quadratic dependency to a linear one and using a group decoding scheme. Also, with the use of sharpness-aware minimization (SAM) optimizer, which minimizes loss value and loss sharpness simultaneously, higher accuracy rates have been reached. In addition, a significant increase in EfficientNet performance is achieved by using image transformations and concatenation together. During the training phase, the random application of the image transformations allows for increasing the image diversity and makes the model more robust. Besides, fusing fundus images of left and right eyes at the pixel level extracts useful information about their relationship. The performance of the final model was evaluated on the publicly available Ocular Disease Intelligent Recognition (ODIR) dataset consisting of 10,000 fundus images, and superior results were obtained in all test set scenarios and performance metrics than state-of-the-art methods. The best results we obtained in the threefold cross-validation scenario for the kappa, F1, and AUC scores are 68.96%, 92.48%, and 94.80%, respectively. Moreover, it can be considered attractive in terms of floating point operations per second (FLOP) and a number of parameters. [ABSTRACT FROM AUTHOR]

Agriculture encompasses a way of life and a profession for the general population. Most global traditions and cultures revolve around agriculture. With the help of advanced farming, agriculture may become more profitable, dependable, and able to use resources and time more effectively. The proposed hybrid CGAN using OTSU and SURF (CGAN-OF) model provides a novel framework for plant leaf disease classification. In proposed CGAN-OF model, contrast-limited adaptive histogram equalization is used for image preprocessing and enhancement. The proposed model utilizes OTSU algorithm to speed up the image segmentation without prior knowledge of the images and SURF algorithm to extract the local features using scale-invariant feature transformation. CGAN increases the input plant village dataset using image generation method and identifies the various plant leaf diseases and classifies them. There are three classifications of leaf diseases: fungi, viruses, and bacteria. Furthermore, these classifications contain approximately 300 diseases. The proposed work identified a minimum of 200 diseases in 18,161 major and minor crop species. The investigational investigations are carried out using the Python Jupyter app with the Kaggle Plant Village Dataset and also leaf samples collected from farmers. The proposed framework achieves 99.2% accuracy. [ABSTRACT FROM AUTHOR]

The success of vision transformer demonstrates that the transformer structure is also suitable for various vision tasks, including high-level classification tasks and low-level dense prediction tasks. Salient object detection (SOD) is a pixel-level dense prediction task that simulates the most salient objects in human visual recognition scenarios. In recent years, depth images have been widely used for salient object detection. Compared with RGB SOD, the key point of RGB-D SOD is the effective fusion of depth information. As RGB-D SOD requires extracting depth features and fusing cross-modal information, additional computation is involved. However, except for lightweight models, most RGB-D SOD methods tend to obtain better prediction maps by consuming more computational resources. We propose a cross-modal dense cooperative fusion net, which provides state-of-the-art performance with less computation and parameters. We take advantage of the ability of the transformer structure to model long sequence dependencies to extract saliency features from RGB images. Since there is less information in the depth image than in the RGB image, it is not necessary to use the same structure in the depth stream. For the sake of reducing parameters and computation, we consider the asymmetric architecture. It is enough to meet our needs that deep features extracted by lightweight MobileV2Net. Our decoder can perform dense cooperative fusion of cross-modal information while decoding features. It can both effectively fuse cross-modal information and save computation. Comprehensive experiments on multiple benchmark datasets for RGB-D SOD show that compared with SOTA methods, our method performs much better with less computation and parameters. [ABSTRACT FROM AUTHOR]

The adversary makes malicious samples capable of triggering erroneous judgments in deep learning models by introducing imperceptible perturbations to the original benign texts. These malicious samples are referred to as adversarial texts. The exploration of adversarial text generation methods not only facilitates our understanding of the robustness of mainstream deep neural networks against such adversarial attacks but also aids in developing appropriate defensive strategies. Nevertheless, the mainstream research on textual adversarial attacks has mainly focused on attack effectiveness, overlooking the associated attack cost. For real-world attacks, considerations such as the time cost, material cost, manpower cost, and various constraints are also crucial. In this paper, we propose a low-cost adversarial text generation method based on the universal strategy in the black-box attack scenario, Universal Chinese Text Tricker (UCTT), for tendency classification on Chinese texts. UCTT is both text-independent and model-independent, which markedly reduces its attack cost. Instead of crafting adversarial texts for a specific text, UCTT generates universal perturbations based on a universal word substitution list, which is applicable to any data in tendency classification datasets. Given a perturbation rate, we can use the word list to craft adversarial texts by simple substitutions without accessing the target model. In the framework of adversarial text generation based on word importance, UCTT utilizes count, arithmetic progression, linear normalization, and nonlinear normalization to calculate the scores of the important words in the dataset and then computes the candidate word frequencies, which in turn constructs the universal word substitution list. Compared with other black-box methods, the experimental results on real-world tendency classification datasets show that UCTT exhibits an effective attack capability while significantly reducing the attack cost. Compared to the powerful baseline we designed that exceeds the SOTA, UCTT improves the efficiency of adversarial text generation by up to a factor of 7 without accessing the target model. In addition to demonstrating excellent attack performance on mainstream models, UCTT is also capable of attacking the powerful ChatGPT in the physical world, which cannot be directly attacked by traditional adversarial text generation methods due to the hard labels produced by the target model. [ABSTRACT FROM AUTHOR]

Cement hydration kinetics, characterized by heat generation in early-age concrete, poses a modeling challenge. This work proposes a physics-informed neural network (PINN) named PINN-CHK designed for cement hydration kinetics, to predict early-age temperature rises in cement paste. PINN-CHK leverages data-driven solutions to craft a high-fidelity prediction model, encompassing material properties and maturity functions in cement hydration. Trained on heated cement paste data, it simultaneously fits experimental results and underlying physics, yielding a mesh-free simulation. Incorporating governing partial differential equations (PDEs), and initial and boundary conditions into its loss function, PINN-CHK architecture undergoes rigorous benchmark testing, demonstrating unparalleled predictive accuracy compared to conventional deep-learning methods. It excels in predicting complete temperature fields during spatial–temporal cement hydration, achieving a remarkable relative L2 error as low as 0.00341. PINN-CHK achieves exceptional convergence and accuracy with only 5% of the training data, ushering in a new era in this crucial field. This innovative approach bridges the gap between theory and practice, offering an attractive alternative to conventional finite element solvers for enhanced comprehension of cement hydration kinetics and concrete maturity and strength development in cement-based materials. [ABSTRACT FROM AUTHOR]

In Aspect-based Sentiment Analysis (ABSA), accurately determining the sentiment polarity of specific aspects within text requires a nuanced understanding of linguistic elements, including syntax. Traditional ABSA approaches, particularly those leveraging attention mechanisms, have shown effectiveness but often fall short in integrating crucial syntax information. Moreover, while some methods employ Graph Neural Networks (GNNs) to extract syntax information, they face significant limitations, such as information loss due to pooling operations. Addressing these challenges, our study proposes a novel ABSA framework that bypasses the constraints of GNNs by directly incorporating syntax-aware insights into the analysis process. Our approach, the Syntax-Informed Attention Mechanism Vector (SIAMV), integrates syntactic distances obtained from dependency trees and part-of-speech (POS) tags into the attention vectors, ensuring a deeper focus on linguistically relevant elements. This not only substantially enhances ABSA accuracy by enriching the attention mechanism but also maintains the integrity of sequential information, a task managed by adopting Long Short-Term Memory (LSTM) networks. The LSTM's inputs, consisting of syntactic distance, POS tags, and the sentence itself, are processed to generate a syntax vector. This vector is then combined with the attention vector, offering a robust model that adeptly captures the nuances of language. Moreover, the sequential processing capability of LSTM ensures minimal information loss across the text by preserving the context and dependencies inherent in the sentence structure, unlike traditional pooling methods. Our experimental findings demonstrate that this innovative combination of SIAMV and LSTM significantly outperforms existing GNN-based ABSA models in accuracy, thereby setting a new standard for sentiment analysis research. By overcoming the traditional reliance on GNNs and their pooling-induced information loss, our method presents a comprehensive model that adeptly captures and analyzes sentiment at the aspect level, marking a significant advancement in the field of ABSA. The syntax distance programming code for required to replicate the experiment is accessible: https://github.com/Makera86/Syntax-Distance.git. [ABSTRACT FROM AUTHOR]

The segmentation of remote sensing images has emerged as a compelling undertaking in computer vision owing to its use in the development of several applications. The U-Net style has been extensively utilized in many picture segmentation applications, yielding remarkable achievements. Nevertheless, the U-Net has several constraints in the context of remote sensing picture segmentation, mostly stemming from the limited scope of the convolution kernels. The transformer is a deep learning model specifically developed for sequence-to-sequence translation. It incorporates a self-attention mechanism to efficiently process many inputs, selectively retaining the relevant information and discarding the irrelevant inputs by adjusting the weights. However, it highlights a constraint in the localization capability caused by the absence of fundamental characteristics. This work presents a novel approach called U-Net–transformer, which combines the U-Net and transformer models for the purpose of remote sensing picture segmentation. The suggested solution surpasses individual models, such as U-Net and transformers, by combining and leveraging their characteristics. Initially, the transformer obtains the overall context by encoding tokenized picture patches derived from the feature maps of the convolutional neural network (CNN). Next, the encoded feature maps undergo upsampling through a decoder and are then merged with the high-resolution feature maps of the CNN model. This enables the localization to be more accurate. The transformer serves as an unconventional encoder for segmenting remote sensing images. It enhances the U-Net model by capturing localized spatial data, hence improving the capacity to capture intricate details. The U-Net–transformer, as suggested, has demonstrated exceptional performance in remote sensing picture segmentation across many benchmark datasets. The given findings demonstrated the efficacy of integrating the U-Net and transformer model for the purpose of segmenting remote sensing images. [ABSTRACT FROM AUTHOR]

Cervical cancer (CC) is affecting women predominantly, and early diagnosis could cure this cancer. This study aims to design and develop an effective deep learning-based classification model to detect early CC stages using clinical data. The proposed method is a combination of an unsupervised deep learning and a supervised neural network, i.e. sparse stacked autoencoder (SSAE) and fuzzy adaptive resonance theory MAP (FAM), respectively, and is denoted as SSAE-FAM. Specifically, SSAE is applied to tackle the data sparsity problem. It extracts the representative features from a data set through feature transformation. The transformed features are then classified by FAM. In this study, a CC data set obtained from the University of California Irvine (UCI) machine learning repository is utilised for evaluation. Owing to missing data in the original CC data set, two data sets are generated from the original CC data samples using two data preprocessing techniques. Both generated CC data sets with four target classes (i.e. Schiller, Cytology, Biopsy, and Hinselmann) are evaluated as four independent binary-class problems. We improve the classification performance of FAM by mitigating the data sparsity problem. Based on a series of experimental studies, SSAE-FAM outperforms other state-of-art methods by achieving 99.47%, 99.34%, 99.48%, and 99.81% mean accuracy rates, respectively, with the first CC data set, and 99.74%, 99.86%, 99.77%, and 99.80% mean accuracy rates, respectively, with the second CC data set. The results positively indicate the usefulness of SSAE-FAM for early CC diagnosis. [ABSTRACT FROM AUTHOR]

Infrared and visible image fusion techniques have emerged as powerful methods to harness the unique advantages of diverse sensors, resulting in improved image quality through the preservation of complementary and redundant information from the original images. While deep learning-based methods have gained widespread adoption in this domain, their predominant reliance on convolutional neural networks and limited utilization of transforms pose certain limitations. Notably, convolutional operations fail to effectively capture long-range dependency between images, which hampers the generation of fused images with optimal complementarities. Because of this, we provide an original end-to-end fusion model built on the swin transform. By modeling long-range dependency using a full-attention feature-encoding backbone, which is a pure transform network with stronger representational capabilities than convolutional neural networks, the model overcomes the shortcomings of manual design as well as complex activity-level measurement and fusion rule design. In addition, we present three loss function strategies that are created expressly to enhance similarity constraints and network parameter training, improving the quality of the detailed information in the fused images. Finally, experimental results on four datasets indicated that our method reached the state-of-the-art in both subjective and objective evaluation compared to eleven state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Artificial neural networks (ANN), machine learning (ML), deep learning (DL), and ensemble learning (EL) are four outstanding approaches that enable algorithms to extract information from data and make predictions or decisions autonomously without the need for direct instructions. ANN, ML, DL, and EL models have found extensive application in predicting geotechnical and geoenvironmental parameters. This research aims to provide a comprehensive assessment of the applications of ANN, ML, DL, and EL in addressing forecasting within the field related to geotechnical engineering, including soil mechanics, foundation engineering, rock mechanics, environmental geotechnics, and transportation geotechnics. Previous studies have not collectively examined all four algorithms—ANN, ML, DL, and EL—and have not explored their advantages and disadvantages in the field of geotechnical engineering. This research aims to categorize and address this gap in the existing literature systematically. An extensive dataset of relevant research studies was gathered from the Web of Science and subjected to an analysis based on their approach, primary focus and objectives, year of publication, geographical distribution, and results. Additionally, this study included a co-occurrence keyword analysis that covered ANN, ML, DL, and EL techniques, systematic reviews, geotechnical engineering, and review articles that the data, sourced from the Scopus database through the Elsevier Journal, were then visualized using VOS Viewer for further examination. The results demonstrated that ANN is widely utilized despite the proven potential of ML, DL, and EL methods in geotechnical engineering due to the need for real-world laboratory data that civil and geotechnical engineers often encounter. However, when it comes to predicting behavior in geotechnical scenarios, EL techniques outperform all three other methods. Additionally, the techniques discussed here assist geotechnical engineering in understanding the benefits and disadvantages of ANN, ML, DL, and EL within the geo techniques area. This understanding enables geotechnical practitioners to select the most suitable techniques for creating a certainty and resilient ecosystem. [ABSTRACT FROM AUTHOR]

Magnetic resonance imaging (MRI) is a medical imaging technique used to show anatomical structures and physiological processes of the human body. Due to limitations like image acquisition time, hardware capabilities, or uncooperative patients, the resolution of MR images is insufficient. Super-resolution (SR) is a crucial method to enhance the resolution of images without expensive scanning equipment. Recent years have witnessed significant progress in MR image super-resolution. Therefore, this survey presents a thorough overview of current developments in deep learning-based MR image super-resolution methods. In general, we can roughly divide the MRI super-resolution methods into single-contrast MR image SR methods and multi-contrast MR image SR methods. Additionally, we introduce the multi-task learning approaches about the MR image super-resolution. We also summarize other crucial topics, such as the degradation model, the definition of the super-resolution problem, the dataset, loss functions, and image quality assessment. Lastly, we indicate the challenges in the field of super-resolution and draw a conclusion to our survey. [ABSTRACT FROM AUTHOR]

Wind power forecasting techniques have been well developed over the last half-century. There has been a large number of research literature as well as review analyses. Over the past 5 decades, considerable advancements have been achieved in wind power forecasting. A large body of research literature has been produced, including review articles that have addressed various aspects of the subject. However, these reviews have predominantly utilized horizontal comparisons and have not conducted a comprehensive analysis of the research that has been undertaken. This survey aims to provide a systematic and analytical review of the technical progress made in wind power forecasting. To accomplish this goal, we conducted a knowledge map analysis of the wind power forecasting literature published in the Web of Science database over the last 2 decades. We examined the collaboration network and development context, analyzed publication volume, citation frequency, journal of publication, author, and institutional influence, and studied co-occurring and bursting keywords to reveal changing research hotspots. These hotspots aim to indicate the progress and challenges of current forecasting technologies, which is of great significance for promoting the development of forecasting technology. Based on our findings, we analyzed commonly used traditional machine learning and advanced deep learning methods in this field, such as classical neural networks, and recent Transformers, and discussed emerging technologies like large language models. We also provide quantitative analysis of the advantages, disadvantages, forecasting accuracy, and computational costs of these methods. Finally, some open research questions and trends related to this topic were discussed, which can help improve the understanding of various power forecasting methods. This survey paper provides valuable insights for wind power engineers. [ABSTRACT FROM AUTHOR]

The inspection and diagnosis of building engineering involve rail surface defect detection, which plays a crucial role in assessing the quality of railway tracks. However, achieving accurate detection remains a significant challenge for infrastructure construction due to the challenging factors, such as complex background, poor texture, the irregular shapes and sizes of rail surface defects, etc. To address these challenges, combined with vision sensor, this paper proposes an automatic vision-based rail surface defect detection network, which leverages a combination of convolutional neural network (CNN) and transformer. Specifically, an effective encoding path based on an improved Res2N-et is presented to obtain stronger contextual information. Meanwhile, to address the limitation of feature representation on global context information, the transformer block is introduced into the bottleneck layer to capture essential global context information. Moreover, an attention-based edge enhancement block is proposed to mitigate the loss of local edge details and strengthen the focus on key feature information, to acquire more discriminant features. To further enrich the feature representation capability, this paper introduces an efficient feature aggregation block, which combines the global features extracted by the transformer and the local features extracted by CNN, to achieve effective feature complementarity and enhances overall detection performance. Combined with public NRSD-MN dataset, proposed model obtains 87.3 % and 77.9 % on PA metric, 85.9 % and 83.4 % in mIoU metric among artificial dataset with natural dataset. Meanwhile, it also obtains 84.0 % PA metric and 87.0 % mIoU metric in SEPR dataset. Experiments have proven the superiority of proposed model though the performance comparison with latest models, which presents an effective and promising rail surface defect detection solution for railway track quality assessment, and can be instrumental in ensuring safe and efficient railway operations. [ABSTRACT FROM AUTHOR]

Clinical decision support systems (CDSSs) can effectively detect illnesses such as breast cancer (BC) using a variety of medical imaging techniques. BC is a key factor contributing to the rise in the death rate among women worldwide. Early detection will lessen its impact, which may motivate patients to have quick surgical therapy. Computer-aided diagnosis (CAD) systems are designed to provide radiologists recommendations to assist them in diagnosing BC. However, it is still restricted and limited, the interpretability cost, time consumption, and complexity of architecture are not considered. These limitations limit their use in healthcare devices. Therefore, we thought of presenting a revolutionary deep learning (DL) architecture based on recurrent and convolutional neural networks called Bi-xBcNet-96. In order to decrease carbon emissions while developing the DL model for medical image analysis and meet the objectives of sustainable artificial intelligence, this study seeks to attain high accuracy at the lowest computing cost. It takes into consideration the various characteristics of the pathological variation of BC disease in mammography images to obtain high detection accuracy. It consists of six stages: identifying the region of interest, detecting spatial features, discovering the effective features of the BC pathological types that have infected nearby cells in a concentrated area, identifying the relationships between distantly infected cells in some BC pathological types, weighing the extracted features, and classifying the mammography image. According to experimental findings, Bi-xBcNet-96 beat other comparable works on the benchmark datasets, attaining a classification accuracy of 98.88% in DDSM dataset, 100% in INbreast dataset with 5.08% and 0.3% improvements over the state-of-the-art methods, respectively. Furthermore, a 95.79% reduction in computing complexity was achieved. [ABSTRACT FROM AUTHOR]

Mental stress is a common problem that affects individuals all over the world. Stress reduces human functionality during routine work and may lead to severe health defects. Early detection of stress is important for preventing diseases and other negative health-related consequences of stress. Several neuroimaging techniques have been utilized to assess mental stress, however, due to its ease of use, robustness, and non-invasiveness, electroencephalography (EEG) is commonly used. This paper aims to fill a knowledge gap by reviewing the different EEG-related deep learning algorithms with a focus on Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks (LSTMs) for the evaluation of mental stress. The review focuses on data representation, individual deep neural network model architectures, hybrid models, and results amongst others. The contributions of the paper address important issues such as data representation and model architectures. Out of all reviewed papers, 67% used CNN, 9% LSTM, and 24% hybrid models. Based on the reviewed literature, we found that dataset size and different representations contributed to the performance of the proposed networks. Raw EEG data produced classification accuracy around 62% while using spectral and topographical representation produced up to 88%. Nevertheless, the roles of generalizability across different deep learning models and individual differences remain key areas of inquiry. The review encourages the exploration of innovative avenues, such as EEG data image representations concurrently with graph convolutional neural networks (GCN), to mitigate the impact of inter-subject variability. This novel approach not only allows us to harmonize structural nuances within the data but also facilitates the integration of temporal dynamics, thereby enabling a more comprehensive assessment of mental stress levels. [ABSTRACT FROM AUTHOR]