Search

The Internet of Medical Things (IoMT) has a network of interconnected medical devices to capture patients' health metrics and store them in a centralized server for analysis by medical experts. The security concerns in IoMT data are therefore very high. The attackers may inject malware into the IoMT data during its transmission. If the health parameters are affected by malware it may mislead the medical experts in making inferences about the patient's health. IoMT devices are resource-constrained and require faster analysis of data for better medical assistance. Most of the classification techniques used earlier suffer from exhaustive time and resource consumption. Hence, developing an intelligent framework that could reduce the data size and classify the malware quickly is essential. This paper proposes a deep learning framework called Auto-encoder to encode the IoMT data. The encoded features are then given to an XGBoost Classifier whose hyperparameters are optimized using the Genetic Algorithm. XGBoost Classifier detects the presence of malware in the IoMT dataset and Clamp dataset with an accuracy of 98.98% and 98.69% respectively. This lightweight model achieves dimensionality reduction through Autoencoder and effectively detects malware through an optimized XGBoost classifier with faster convergence and limited computational costs. Implementing this framework on the fog layer provides promising results by optimizing resource usage and saving time compared to previous works. Moreover, the proposed method does not need explicit feature selection from the dataset when making predictions. [ABSTRACT FROM AUTHOR]

Visual tracking is a fundamental task in computer vision, which extracts the target context descriptions and features from the first image frame and tracks the target in subsequent frames with the updated target appearance models and the updated tracking models accordingly. The existing deficiencies of deep learning based visual tracker are: precision-speed dilemma, inadequate context usage and spawning tracking noises. To achieve real-time and accurate tracking with deep learning modules under tracking-by-detection framework, this paper introduced a discriminative target predictor based on temporal-scene attention context enhancement and candidate matching mechanism (ACDP). The contributions of ACDP include: 1) a vision transformer based temporal context enhancer is suggested by formulating the temporal context enhanced feature extractor via multi-head attention mechanism; 2) a target state propagation scheme based scene context enhancer is proposed via constructing the target state matrix and corresponding state forward propagation techniques; 3) a joint prediction method is provided to take full usage of the enhanced context through candidate selection and matching mechanism based on dust-bin conception network; 4) theoretical proofs for vision transformer acceleration scheme and the error bounds under successive tracking failure circumstances are given. In addition, the extensive experiments under OTB100, UAV123, NFS, AVisT and VOT2018 benchmarks evaluate the favorable performances of ACDP and indicate that: 1) the proposed modules in ACDP can achieve comparable tracking results even under difficult sequences and scenarios against other 21 state-of-the-art trackers; 2) the ACDP performs 35 FPS tracking speed on the 5 benchmarks averagely. [ABSTRACT FROM AUTHOR]

The human skin, the largest organ with multiple layered functionalities, houses melanocytes in the deeper strata of its epidermis. These cells can be adversely impacted by ultraviolet radiation, thereby instigating melanoma, the deadliest form of skin cancer. Failure to detect melanoma at an early stage can potentially lead to metastasis, forming complex tumors in other tissues. Despite substantial efforts, visual inspections can occasionally overlook melanoma cases due to inherent subjectivity. To surmount this challenge, an automated detection system is necessary. Recent attempts to establish such a system have predominantly employed "push-through" strategies involving deep (neural) networks and their ensembles, which however necessitate significant computational resources. This paper presents a novel approach, amalgamating a conventional machine learning technique, Bag of Visual Words, with a pretrained deep neural network for comprehensive deep feature extraction from enhanced input image patches. The proposed method, assessed on the ISIC Challenge 2017 dataset, surpassed all other entries on the challenge leader-board, registering an accuracy of 96.2% in the task of lesion classification. • Novel method for melanoma detection combining Bag of Visual Words with a deep network. • Patch-based feature extraction from lesion images integrated into BoVW with K-means. • Successful image enhancement technique for feature extraction from lesion images. • Efficient framework: modular architecture, easy updates, less training delay. [ABSTRACT FROM AUTHOR]

A metaheuristic-based structural optimization method, whilst being popularly adopted due to its advantages in by-passing gradient function calculations, requires the use of time-consuming advanced analyses for constraint evaluation. To overcome this drawback, the present paper proposes a robust (machine learning-based) optimization method that combines the light gradient boosting machine (LightGBM) with the efficient p-best differential evolution (E p DE) method. In essence, the LightGBM classification model is constructed to assess the constraint (safety and integrity) satisfaction of structures. An efficient framework using a so-called safety parameter is proposed to prevent inaccurate predictions of the LightGBM model. The E p DE processes the optimization procedures on the constructed classification LightGBM model. This enables an enhanced machine learning-based optimization technique that not only maintains the sufficiently accurate optimal design of structures but also significantly reduces the required computing efforts, as compared to standard optimization schemes. Various examples of steel structure designs (i.e., two of which have been provided herein) have been successfully performed by the proposed approach. These illustrate the accuracy and robustness of the proposed method, where good comparisons with reference algorithms (including standard DE with "DE/rand/1" mutational strategy, Jaya, Rao-1 and CaDE) are evidenced. The statistical values collected present the accuracy and reliability of the proposed method in obtaining the minimum total weight designs with substantial (some 60%) reduction of computing efforts required to furnish optimal results. [ABSTRACT FROM AUTHOR]

Stock price forecasting has been always a difficult and crucial undertaking in the field of finance. In the last few decades, deep learning models based on RNNs and LSTMs have dominated the research, where the stock price data are modeled as time series data. However, the high volatility of stock prices and the decay of information learned from historical data prevented these models from achieving more accurate predictions in this problem. Recently, Transformer has been gradually applied in time series prediction, but the methods aim to feed the highly-uncertain social media information as the additional auxiliary information into Transformer, rather than improving the ability to extract features from historical series. In this paper, we propose a Series Decomposition Transformer with Period-correlation (SDTP), which uses the period-correlation mechanism and series decomposition layers to further discover relation between historical series and learn the changing trends in the stock market for high forecasting accuracy and generalizability. The extensive experimental results show that the proposed SDTP model generally outperforms the state-of-the-art methods on a collection of datasets. • A novel Transformer-based model called SDTP is proposed. • Stock series is decomposed into trend and season parts. • SDTP forecasts the stock index without highly-uncertain social text. • SDTP has higher prediction accuracy compared with the existing methods. [ABSTRACT FROM AUTHOR]

Recent studies focus on the utilization of deep learning approaches to recognize depression from facial videos. However, these approaches have been hindered by their limited performance, which can be attributed to the inadequate consideration of global spatial–temporal relationships in significant local regions within faces. In this paper, we propose Spatial–Temporal Attention Depression Recognition Network (STA-DRN) for depression recognition to enhance feature extraction and increase the relevance of depression recognition by capturing the global and local spatial–temporal information. Our proposed approach includes a novel Spatial–Temporal Attention (STA) mechanism, which generates spatial and temporal attention vectors to capture the global and local spatial–temporal relationships of features. To the best of our knowledge, this is the first attempt to incorporate pixel-wise STA mechanisms for depression recognition based on 3D video analysis. Additionally, we propose an attention vector-wise fusion strategy in the STA module, which combines information from both spatial and temporal domains. We then design the STA-DRN by stacking STA modules ResNet-style. The experimental results on AVEC 2013 and AVEC 2014 show that our method achieves competitive performance, with mean absolute error/root mean square error (MAE/RMSE) scores of 6.15/7.98 and 6.00/7.75, respectively. Moreover, visualization analysis demonstrates that the STA-DRN responds significantly in specific locations related to depression. The code is available at: https://github.com/divertingPan/STA-DRN. [ABSTRACT FROM AUTHOR]

Deep learning (DL) has made its way into many disciplines ranging from health care to self-driving cars. In financial markets, we see a rich literature for DL applications. Particularly, investors require robust algorithms that can navigate and make sense of extremely noisy and volatile markets. In this work, we use deep learning to select a portfolio of stocks and use a genetic algorithm to optimize the hyperparameters of DL. The work analyzes the improvement in using genetic-based hyperparameter optimization over grid searches. The Genetic Algorithm brings 40% improvements in prediction when compared to a random-grid search. Novelty-wise, the work couples a genetic-based hyperparameter optimization with multiple Deep RankNet models to predict the behavior of financial assets. Our results show promising portfolio returns 20% better than the general market. In the highly volatile COVID 19 period, the models exceed market returns by more than double. Overall, this paper brings a comprehensive work that integrates hyperparameter optimization, Deep RankNet, LSTM, period size variations, input variable transformation, feature selection, training/evaluation ratio analysis, and multiple portfolio selection strategies. [ABSTRACT FROM AUTHOR]

Compared with the traditional vibration displacement measurement methods, visual vibration measurement offers several advantages such as long-distance capability, non-contact operation and easy installation. However, the phenomenon of low fitting accuracy of the bounding box often occurs when detecting rotating objects, resulting in a slight deviation in the relative offset of the center vibration point of the target between frames, which will cause a serious deviation in the regression of vibration displacement offset. In this paper, a high-speed industrial camera is employed as the image acquisition medium, and a deep learning-based semantic segmentation method is introduced to address visual vibration measurement challenges in rotating body. Specifically, the CSP module integrates different depth semantic information which is introduced into the Mobiledets backbone network in a targeted manner. This is not only strengthens the performance of the network for segmenting vibration objects, but also dramatically improves the practical performance of the algorithm. The conventional Relu activation function is substituted with Mish activation function, making the network more adept at segmenting rotating body in challenging backgrounds with varying illumination, blur, and similarity. The CSP+Mobiledets backbone network constructed in this study outperforms the U-Net network in terms of feature extraction effectiveness. Adding Dice-loss to the original loss function can more effectively solve the severe imbalance problem of samples caused by long-distance image acquisition. We take the most representative rotating body-rotor as the experimental subject. The displacement curve obtained by the existing algorithm has the best degree of fit with the signal curve collected by the eddy sensor. The results of different segmentation algorithms and detection algorithms on time domain curve plot, frequency domain plot and axis orbit plot are collectively compared. Furthermore, the results also provide valuable guidance for visual measurement of the vibration displacement of the rotating body in specific industrial scenarios. • A semantic segmentation network is applied for vibration displacement measurement. • CSP+Mobiledets backbone network to enhance feature extraction are proposed. • The proposed network has excellent performance under complex background. • The proposed algorithm has broad applicability on rotating structure. [ABSTRACT FROM AUTHOR]

Multi-focus image fusion techniques are designed to integrate the focused regions of multiple source images to produce an all-in-focus image. Unfortunately, the fusion results generated by existing methods are prone to suffer from chromatic aberrations and blurred boundary regions. In this paper, a multi-focus image fusion framework combining focus property detection and deep image matting is proposed to achieve high-quality fusion results. Specifically, we introduce spatial frequency detection blocks, which integrate the spatial frequency information of source images at different levels to achieve effective focus property detection for the initial decision. Then, the trimap acquired from the initial decision is optimized using a deep matting model so as to further detect the focused/defocused boundaries and obtain a more precise decision map. Meanwhile, a new hybrid loss function is introduced, and a large-scale dataset is created to improve the performance of the fusion framework. Finally, the fusion result is obtained by integrating the focused regions via the decision map. Compared to 18 other advanced fusion methods, the proposed method outperforms all of them in terms of the seven objective metrics evaluated on two publicly available datasets. These experimental results provide strong evidence that the proposed fusion framework excels not only in subjective vision but also in quantitative analysis. The code of the proposed method will be available at https://github.com/govenda/DIMF. • The first application of deep image matting to multi-focus image fusion community. • Proposed a new focus property detection approach. • Presented a hybrid loss function and created a large-scale dataset. • Solved the problem of blurred decision map boundary. [ABSTRACT FROM AUTHOR]

High-frequency financial data is more difficult to predict than low-frequency data because it possesses nonlinearity, nonstationarity, higher volatility, and long memory and is frequently accompanied by the jump phenomena. In this paper, the nonparametric regression (NR) model based on kernel function is used to fit the nonlinear relationship between the nonstationary series Y t and its lagging series to model the trend of high frequency financial time series. Furthermore, the deep learning LSTM (long short-term memory) model is applied to capture the high volatility and frequent jumps of high frequency financial data and to improve the forecasting accuracy of the residual series. The results demonstrate that the hybrid NR and LSTM model has greatly improved the forecasting accuracy in several evaluation criteria. In comparison to NR, support vector machine (SVM), LSTM, ARIMA and NR-SVM models, the mean absolute error (MAE) of NR-LSTM has reduced by 89.78%, 97.85%, 86.48%, 32.47% and 89%, respectively. In addition, we have constructed the trading strategy for the Shanghai-Shenzhen 300 index by using the NR-LSTM model. The NR-LSTM model can continue to provide good returns even during a bear market, which can serve as a guide for investors. Furthermore, the NR-LSTM model also exhibits the best forecasting effect when we model the high-frequency data of Ping An bank in China, the FTSE 100 index in the UK, and the S&P 500 index in the US. [ABSTRACT FROM AUTHOR]

Ultrasound assessment of fetal biometry is widely used for gestational age prediction and abnormal fetal growth diagnosis. However, manual biometric measurements require time-consuming procedures and current automated measurement approaches remain challenging in terms of accuracy and computational complexity. This paper proposes deep learning-based efficient automatic fetal biometry measurement method for system-on-chip (SoC) solution. For end-to-end automated measurements, a hardware-friendly bilateral segmentation network (H-BiSeNet) was designed and optimized for multiple fetal objects (i.e., head, abdomen and femur), and fetal biometric parameters were then calculated with a robust multiple keypoint detection algorithm. For hardware implementation with a SoC, a deep learning processing unit (DPU) was employed to accelerate the proposed segmentation network, and biometric measurements with image pre- and postprocessing were conducted on an application processing unit (APU). A total of 6000 fetal ultrasound images in the three regions were collected from 2568 subjects, and the dataset was utilized to train the proposed network. In the performance evaluation, the proposed segmentation network with optimal quantization (H-BiSeNet-Q) outperformed other segmentation networks (i.e., U-Net, U-Net++, Attention U-Net, U-Net3+, DeepLabv3 + and BiSeNetv2) in terms of the average Dice coefficient, and automated measurements with H-BiSeNet-Q were highly correlated with manual measurements (r ≥ 0.983, p < 0.01 for all parameters, i.e., biparietal diameter, head circumference, abdominal circumference and femur length). In the execution time analysis, H-BiSeNet-Q ran at 4.25 frames per second (FPS) with parallelism corresponding to 80% resource utilization in the SoC. These results showed feasibility in low-resource hardware settings such as portable ultrasound systems. [ABSTRACT FROM AUTHOR]

AI-powered analysis of histopathology data has become an invaluable assistant for pathologists due to its efficiency and accuracy. However, existing deep learning methods still face some challenges in specifying cancer subtypes. For example, the ultra-high resolution of histopathological slides generally contains numerous redundant features, which are not useful for cancer subtype classification and thus lead to considerable computational costs. Moreover, the lack of manual annotations of disease-specific regions (i.e., patch-level annotations) from experts makes it more difficult to learn such histological features with only slide-level labels. In this paper, we propose an end-to-end weakly supervised learning framework called EWSLF to address these issues. First, we employ a cluster-based sampling strategy to refine the histological features for further training, which can improve classification accuracy and reduce computational cost. Second, we employ a multi-branch attention mechanism to produce patch-level pseudo-labels and aggregate the patch features into slide-level features, which can complement the missing patch-level labels from experts. Experimental results on both public and in-house datasets demonstrate the superiority and credible results of our model compared with the state-of-the-art methods for cancer subtype classification. Code: https://github.com/hongren21/ewslf. [ABSTRACT FROM AUTHOR]

In the context of a resilient energy system, accurate residential load forecasting has become a non-trivial requirement for ensuring effective management and planning strategy/policy development. Due to the highly stochastic nature of energy load profiles, it is difficult to predict accurately, and usually, predictions are error-prone. This paper explores the potential of Empirical Mode Decomposition (EMD) in simplifying the dynamics of complex demand profiles. The simplified components are then embedded within a deep learning model, specifically Convolution Neural Network (CNN) and Long Short-Term Memory (LSTM), to forecast short-term residential loads. The novel modelling framework integrates Bayesian optimisation strategy, feature decomposition technique, feature engineering phase, and percentile-based bias correction algorithm to enhance model accuracy. The model is developed using a case-study residential dwelling located in Fintry (Scotland), and the model performance is assessed over four forecast horizons. The overall efficiency of framework is also investigated for three algorithms: random forest, gradient boosting decision trees (GBDT), and an LSTM network. While EMD and feature engineering were found to greatly improve prediction accuracy, the number of IMFs used was shown to significantly impact the model's performance and computational complexity. The model was tested on two further case studies from Fintry. • A novel forecasting framework is proposed for short-term energy demand prediction. • Multi-domain features are extracted to increase the accuracy. • Utilising empirical mode decomposition to enhance framework performance. • Exhibit strong forecasting capability without the additional computational cost. • Excellent forecasting accuracy and model robustness are validated using three households. [ABSTRACT FROM AUTHOR]

The Computational Fluid Dynamics (CFD) simulations at different fidelity levels are a common tool for predicting the aerodynamic performance in many engineering applications such as; automotive and aerospace to cite a few. However, the associated computational costs with these simulations are not trivial, and using the CFD simulations might easily become unpractical specially at the early design stages. The number of the required CFD simulations could be reduced immensely by using Artificial Intelligence (AI) methods which can predict, for example, the aerodynamic coefficients based on previous and limited CFD simulations. Having discussed the AI potential to minimize the CFD usage, the objective of this work is to investigate and compare the ability of different AI methods to predict the aerodynamic coefficients using the same CFD database from the literature. In this study, three different AI methods, namely; machine learning (e.g. Extreme Gradient Boosting (XGB), CatBoost, Bagging, Light Gradient Boosting Machine (Light-GBM), Random Forest (RFR), Gradient Boosting), deep learning (e.g. One Dimensional Convolutional Neural Network (Conv1D)) and surrogate models (e.g. Gaussian Process Regression (GPR)) have been utilized to estimate a sample CFD database (e.g. NACA0012, RAE2822) from the existing literature, and the prediction performance of these methods have been compared. The results show that, the aerodynamic coefficients predicted using CatBoost, XGB and Bagging are in good agreement with the reference CFD results. Furthermore, the proposed Conv1D method with LeakyReLU activation function showed a promising results in the development of aerodynamic models which can be used in the early stages of the air vehicles design. The obtained results from this work indicate that the CatBoost and the XGB models required much smaller training time compared to the Conv1D method, moreover, the proposed regression trees; CatBoost, Bagging and XGB can reduce the required number of CFD simulations significantly. • Various artificial intelligence methods were studied to predict aerodynamic data. • Regression tree models such as XGB, Catboost and LightGBM were used. • Surrogate models and deep learning models (i.e. Conv1D) also were used. • Detailed comparison were presented between the aforementioned models. • The paper suggests the suitability of XGB and Catboost for aerodynamics cases. [ABSTRACT FROM AUTHOR]

Aquaculture net pen defects such as biofouling, vegetation, and holes are key challenges to efficient and sustainable fish production in aquaculture. These defects must be monitored to avoid problems with aquaculture net safety as well as fish growth. Recently, deep learning methods have been adopted to solve detection and classification problems in different applications. However, the conventional methods are challenging to meet the demands of high precision and real-time detection of aquaculture net defects in a complex marine environment. Towards this end, this paper proposes an autonomous net pen defect detection system that contains a novel multi-scale semantic segmentation topology for detecting the biofouling, vegetation, and hole problems in the aquaculture environment. In particular, we emphasize fusing the attention maps obtained across different decomposition levels of the network to generate rich feature distributions that enable the accurate extraction of biofouling, vegetation, and holes within aquaculture nets. Moreover, the proposed model is thoroughly tested on two private datasets and two public datasets which were acquired from a real-field fish farms. Across all four datasets, the proposed framework showed remarkable biofouling, vegetation, and hole detection performance, where it outperformed state-of-the-art methods by 6.58%, 3.69%, 6.44%, and 4.78% in terms of mean average precision across LABUST, KU, NDv1, and NDv2 datasets, respectively. • Detection of aquaculture net defects such as biofouling, vegetation, and net holes. • A novel multi-scale encoder–decoder topology. • A new loss function, dubbed L t , can constrain the segmentation networks. • A release of the two highly annotated aquaculture datasets. [ABSTRACT FROM AUTHOR]

Identification of human malicious behaviors is a demanding problem in nuclear security due to its potential to cause huge personnel and economic casualties. With the visibly rapid development of deep learning technology especially computer vision, the automatic identification of human malicious behaviors become feasible, offering significant savings in human resources. However, there is few research focusing on the explainable reasoning process based on the calculation results of computer vision. In this paper, an innovative framework is proposed, which contains two modules called computer vision module and reasoning module. Among them, the computer vision module could extract and detect human actions by some computer vision technologies, while these actions would be comprehensively utilized in reasoning module to derive the reasoning results. Furthermore, within the reasoning module, three novel reasoning methods are developed which are data-based reasoning method, language-based reasoning method and graph-based reasoning method. For evaluating the practicality and effectiveness of these methods, experiments are conducted on four typical scenarios in nuclear security, which are normal status, fence climbing, wire cutting and weapon holding. The results show that the proposed language-based reasoning method turns out to be the best one, which obtain a higher precision value of 0.7917 and a perfect recall value of 1.0000. [ABSTRACT FROM AUTHOR]

Efficient and accurate medical image segmentation is necessary for pathological evaluation and disease diagnosis in clinical practice. In recent years, the U-shaped encoder–decoder structure has achieved good performance in various medical image segmentation tasks. It is a neural network model similar to the letter U, which can gradually reduce the spatial dimension of the feature map to capture high-level semantic information in the feature encoding stage, and can make up for the lack of target detail information in the feature decoding stage to decouple the final segmentation results. However, the U-shaped encoder–decoder structure still suffers from semantic asymmetry and global semantic dilution problems, which are exacerbated during decoding, due to the limitation of the framework's fixation. In this paper, we propose a cross-level collaborative context-aware framework (C 3 -Net) to address the aforementioned issues in medical image segmentation. The main contributions of this research include: (i) To address the inherent problems in the U-shaped structure, we propose the C 3 -Net to explore the differences between cross-level contextual information effectively; (ii) To achieve detailed information preservation and semantic information enhancement, we present a Channel Scale-aware Context Enhancement (CSCE) module to enhance low-level contextual features from the global and local scales of the channel dimension; (iii) A Spatial Pyramid Context Alignment (SPCA) module is designed to extract and align pyramid features, thus obtaining accurate global contextual features at the network's top; (iv) We propose a Cross-Level Collaborative Context Refinement (CCCR) module to enable efficient collaboration between cross-level contextual features to simultaneously achieve semantic alignment and global semantic enhancement. The outcomes of implementing the proposed method on three publicly available datasets (Synapse, ACDC, and GlaS) and one private dataset (VMICH) show that the proposed method outperforms all competing approaches on four datasets. Our C 3 -Net achieves new state-of-the-art performance on four image segmentation tasks with Dice scores of 85.26%, 92.10%, 87.20%, and 87.61%, respectively. • We propose a novel framework called C3-Net for medical image segmentation tasks. • Our C3-Net address the inherent semantic gap and global semantic dilution problems. • We evaluate the effectiveness of C3-Net on four public and private datasets. • The results show that our C3-Net outperforms other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Accurate solar irradiance forecasting with fine spatiotemporal correlations is essential for photovoltaic power generation. However, current spatiotemporal methods are unable to extract spatial features appropriately, leading to rough prediction results. Consequently, a deep graph neural network is proposed in this paper for multi-site solar irradiance forecasting. Self-adaptive adjacency matrices were developed to better capture the intrinsic influence between sites under different climate types at various time steps. Based on the adaptive graph convolution and long short-term memory network (AGCLSTM), the spatiotemporal features of each site were extracted to obtain prediction results for all the sites in the research area simultaneously. In the experiments conducted on the datasets of three climate types, the average root mean square error (RMSE), mean absolute error (MAE), and correlation coefficient (R) of the proposed model are observed to be in a range of 47.519–54.854 W/m2, 18.968–24.426 W/m2, and 0.9737–0.9889, respectively. Compared to the benchmark models, the average RMSE and MAE reduce by 1.31%–10.86% and 3.75%–18.57%, respectively. Moreover, the proposed method proves to be effective for scenes of solar irradiance fluctuations with the help of self-adaptive adjacency matrices, which enhances its generalization ability and prediction accuracy for one-hour-ahead solar irradiance forecasting. [ABSTRACT FROM AUTHOR]

While searching for consumer goods, users frequently ask for suggestions from their peers by writing short free-form textual requests. For example, when searching for movies users may ask for " Drama movies with a mind-bending story and a surprise ending, such as Fight Club " in one of the many online discussion boards. Despite the recent developments in large language models (LLMs) and natural language processing (NLP), modern recommender systems still struggle to process such requests. Therefore, in this paper we evaluate several approaches for annotating structured information from such short, free-form natural language user texts to calculate recommendations. We set up this evaluation as a two phase processes including (a) identification of the best NLP approach to identify key elements of users' requests, and (b) assessment of the quality of recommendations computed with such elements. For our evaluation, we use a gold-standard reddit movie recommendation dataset consisting of annotations, manually created by crowdworkers who extracted keywords, actor names and movie titles. Using this dataset we evaluate a collection of more than 30 NLP and five recommender approaches. In addition, we perform an ablation study to assess relative annotation importance for movie recommendations. We find that domain-specific deep learning models, trained on a subset of data as well as embedding-based recommendation approaches are able to match the recommendation performance of recommendations computed from manual annotations. These promising results warrant further investigation in automatic processing of short free-form texts for computation of recommendations. Specifically, we provide insights into which NLP models and configurations work best for automatically annotating free text to compute (movie) recommendations, hence substantially reducing the search space for combinations of NLP and recommendation algorithms in the movie and potentially other domains. • Deep learning models in combination with external embedding techniques work best. • Rec. performance based on NLP matches the performance based on manual annotations. • Sentiment of NLP annotations does not significantly improve recommender performance. • Movie titles and movie keywords are the most important annotations for accurate rec. [ABSTRACT FROM AUTHOR]

The COVID-19 pandemic's profound impact on global health and the economy underscores the need for new computational strategies to rapidly identify potential antiviral drugs against emerging and re-emerging infectious diseases. AI-based drug repurposing methods have the potential to significantly shorten the drug discovery process, rendering them a crucial and efficacious approach for screening anti-SARS-CoV-2 compounds. In this paper, we proposed GINCM-DTA, an AI-based graph isomorphism network with protein contact map representation for drug–target affinity (DTA) prediction, and presented the COVID-DTA dataset for SARS-CoV-2-specific drug repurposing. Pre-training GINCM-DTA on large-scale Davis and KIBA datasets achieved state-of-the-art performance, while fine-tuning on the COVID-DTA dataset using transfer learning resulted in a mean square error of 0.026 and concordance index of 0.967. Utilizing the fine-tuned GINCM-DTA model, we predicted DTAs for drugs against SARS-CoV-2 main targets, focusing on key conserved proteins and unmutated spike proteins for virtual screening. We integrated GINCM-DTA predictions with molecular docking simulations to identify five candidates. In vitro assays revealed Hydralazine (5.737 nM) as a potential inhibitor of ACE2 and TMPRSS2, effectively blocking SARS-CoV-2 infection. Moreover, the results highlight the potential of Hydralazine as a broad-spectrum drug candidate, demonstrating its efficacy in obstructing the binding between ACE2 and spike proteins of multiple Omicron variants, including the currently dominant strains BQ.1, BQ.1.1, XBB.1.5, and XBB.1.16. Our method offers a novel computational strategy for COVID-19 antiviral screening and prepares for potential future epidemics incited by emerging viruses. • Proposed GINCM-DTA: a novel GIN model for DTA prediction with SOTA performance. • The fine-tuned GINCM-DTA achieved excellent performance: MSE=0.026, CI=0.967. • Identified Hydralazine as a potential inhibitor to block SARS-CoV-2 infection. • Potential effective against Omicron subvariants: BQ.1, BQ.1.1, XBB.1.5, XBB.1.16. • New computational strategy for COVID-19 antiviral screening & future virus epidemics. [ABSTRACT FROM AUTHOR]

Nowadays, users across the globe interact with one another for information exchange, communication, and association on various online social media. However, some individuals exploit these venues for malicious practices like hate speech and cyberbully. In this paper, we present an improved multilingual hate speech and cyberbully detection model using bagging-stacking based hybrid ensemble deep learning techniques. The proposed model utilizes Bi-directional Long Short-Term Memory (BiLSTM), Bi-directional Gated Recurrent Unit (Bi-GRU), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) techniques to enhance the overall performance. We first preprocess the multilingual data streams followed by adoption of Global vectors for word Representation (GloVe) embeddings to convert words to a vector representation in parallel enabling the data streams for binary classification task. In order to construct an architecture for the detection of hate speech and cyberbully, we introduce a heterogeneous fusion of multiple effective models in a unique approach such that CNN-LSTM utilizes a stacking approach with stochastic gradient descent to achieve optimal weights, whereas all the base learners used bagging ensemble approach with cross-validation to reach optimal weights. The final output layer of the proposed ensemble deep learning architecture is achieved using a super learner approach on base learners. To show the efficacy of the proposed model, we conduct the simulation on a total of nine real-world social media datasets in different languages and compared the results with other contemporary hate speech and cyberbully detection methods. The collected findings show that the proposed model outperforms other models on considered datasets and shows an improvement of at least 4.44% in F1 scores. • Proposed an ensemble DL multilingual model for hate speech & cyberbully detection. • Model combines bagging, stacking, and super learning methods to improved results. • Model adopts a light weight heterogeneous fusion and bagging-stacking hybrid methods. • The experimental findings advocate the superiority of the proposed model. [ABSTRACT FROM AUTHOR]

Radar high-resolution range profile (HRRP) plays a crucial role in ballistic target recognition due to its simplicity and fast computation. Convolutional neural networks (CNNs), a popular deep learning approach, are widely used in radar automatic target recognition (RATR) based on HRRP, thanks to their excellent feature extraction capabilities. However, current research mainly focuses on integrating branch networks like recurrent neural networks and attention mechanisms into vanilla CNNs to enhance feature extraction. Limited research exists on reducing the computational complexity of vanilla CNNs when applied to HRRP-based RATR, as well as target recognition problems with limited samples. To address these issues, this paper proposes a lightweight depth-wise separable fusion CNN (DSFCNN) for ballistic target HRRP recognition. The DSFCNN reduces the computational complexity of vanilla CNNs while improving recognition accuracy. Furthermore, we introduce the sample-level fitting and class-level distinguishability (SFCD) loss for limited-sample ballistic target HRRP recognition. The SFCD loss ensures better fitting performance of all samples, as well as higher intra-class compactness and inter-class separability, enhancing the distinguishability of embedded features in limited-sample conditions. During the training phase, the SFCD loss is equally applied to four DSFCNNs with shared structures and parameters, resulting in a model known as Quadruplet DSFCNN (QDSFCNN). However, during the testing phase, only one DSFCNN is utilized. Additionally, we propose the parameter quantity shifting-fitting performance (PQS-FP) coordinate system to elucidate the performance disparities between DSFCNN and vanilla CNNs in our experiments. PQS-FP offers inspirational insights into the relations among parameter quantity, fitting performance, and recognition performance. Extensive experimental results demonstrate the robustness and effectiveness of QDSFCNN for ballistic target recognition in various limited-sample scenarios. [ABSTRACT FROM AUTHOR]

Early rainfall prediction is very important to ensure the economic balance of any agriculture-dominated country, such as India. Deep learning has recently received considerable attention for predicting rainfall using historical data. However, rainfall depends on multiple factors like ElNino, rising sea levels, warming of the Atlantic Ocean, etc. The changing climate has also significantly increased rainfall variability. The current deep-learning approaches are either based on a limited set of predictors or do not utilize the time-series properties for the recurrence of similar climate conditions every 12 months. Besides, they perform poorly in predicting accurate rainfall estimates for the diverse areas of India due to limited data availability across various regions. Therefore, this paper presents a groundbreaking solution to address the challenge of limited data by introducing RfGan (Rainfall GAN), a novel variant of generative adversarial networks. RfGan aims to fulfill the crucial need for abundant data to ensure the effectiveness of model training. Moreover, a novel rainfall prediction method, RfGanNet, has been introduced that leverages the strengths of the proposed RfGan for data augmentation, long-short-term memory, and convolutional neural networks for rainfall prediction. The performance of RfGanNet has been validated for India and its clustered regions in terms of nine parameters: precision, recall, specificity, Matthews correlation coefficient, accuracy, F1-score, mean absolute error, mean square error and computation time. The experimental results of RfGan are compared against original and synthetic datasets generated from state-of-the-art methods, namely tabular gan (TGAN), Wasserstein gan (WGAN), and conditional tabular gan (CTGAN). The experimental results indicate that the proposed RfNet model, when trained using RfGan-generated data, surpassed all the other considered methods by achieving 97.22 % accuracy. Moreover, RfNet has been compared with other related deep Learning approaches to validate its efficacy. Additionally, an ablation study investigating various LSTM–CNN combinations has been conducted. The findings of the study indicate that oceanic and climate-related predictors have a noteworthy influence on Indian rainfall. Moreover, the study highlights the suitability of combining Bidirectional LSTM, CNN, and multiple predictors for accurate rainfall prediction. • Novel variant of generative adversarial network- RfGan is proposed. • Different combinations of LSTM and CNN for rainfall prediction are tested. • Novel classifier — RfNet is tested on different augmented datasets. • K-means clustered regions of India have been identified. [ABSTRACT FROM AUTHOR]

Modifying some keywords or numbers on documents to change the original intention is illegal. In some litigation cases, especially economic cases, there is often a need to examine the type of ink on documents. This paper proposes a nondestructive and accurate detection method for ink inspection. In this work, a high-information interactive, full group residual convolution network (FGRC-Net) is proposed and combined with spectral information to classify the inks of six different brands of pens. First, FGRC-Net realizes the information interaction among channels to effectively improve the feature extraction ability through multigroup convolution and multipath cascade. Meanwhile, to avoid feature degradation, the residual connection is introduced. Second, the ink spectral information of six pens based on the hyperspectral system is obtained. Finally, in comparing FGRC-Net and other deep learning methods, FGRC-Net shows the best classification performance with 98.38% accuracy, 98.53% precision, and 98.40% recall. The results show that combining FGRC-Net with a hyperspectral system is an effective detection method for handwritten ink classification. It also provides an effective detection method for handwritten document ink inspection. At the same time, the universality and validity of FGRC-Net in processing spectral information are verified by the comparison of multiple datasets. [ABSTRACT FROM AUTHOR]

Automatic pollen detection based on light microscope (LM) images is helpful for pollinosis symptoms prevention. Recently, many deep learning methods have been proposed to identify pollen grains based on multi-scale feature fusion mechanism. However, in real scenarios, there are two main challenges that need to be considered: (1) Complex pollen characteristics; (2) Irrelevant objects interference. It means that the pollen detection requires not only learning the relationship among multi-scale features but also refining the feature representation. To this end, this paper proposes an attention-based multi-scale feature fusion network (AMFF-Net) for automatic pollen detection on real-world LM images. The proposed AMFF-Net includes three modules: The feature extraction module utilizes the series-connection attention to capture the spatial and channel dependencies of different level feature maps (for solving the challenge 1). In the feature fusion module, a parallel connection attention is able to learn more discriminative feature representation based on bidirectional pathway guidance (for solving the challenge 2). Both are jointly adopted to enhance the representational capacity of the final results in the pollen prediction module. Extensive experiments are conducted on the real-world RPD dataset, and our AMFF-Net achieves the best performance (83.9% of mean average precision) comparing with other state-of-the-art methods. We believe that this work can serve as an important reference for the development of pollen monitoring system in a real scenario. • An attention-based feature fusion network is proposed for pollen detection. • Series connection attention aims to learn multiple dependencies of feature maps. • Parallel connection attention is used to extract discriminative features. • Both are jointly adopted to enhance the representational capacity. • Extensive Experiments are conducted to prove the effectiveness of our method. [ABSTRACT FROM AUTHOR]

In implant prosthesis treatment, the design of the surgical guide heavily relies on the manual location of the implant position, which is subjective and prone to doctor's experiences. When deep learning based methods has started to be applied to address this problem, the space between teeth are various and some of them might present similar texture characteristic with the actual implant region. Both problems make a big challenge for the implant position prediction. In this paper, we develop a two-stream implant position regression framework (TSIPR), which consists of an implant region detector (IRD) and a multi-scale patch embedding regression network (MSPENet), to address this issue. For the training of IRD, we extend the original annotation to provide additional supervisory information, which contains much more rich characteristic and do not introduce extra labeling costs. A multi-scale patch embedding module is designed for the MSPENet to adaptively extract features from the images with various tooth spacing. The global–local feature interaction block is designed to build the encoder of MSPENet, which combines the transformer and convolution for enriched feature representation. During inference, the RoI mask extracted from the IRD is used to refine the prediction results of the MSPENet. Extensive experiments on a dental implant dataset through five-fold cross-validation demonstrated that the proposed TSIPR achieves superior performance than existing methods. [ABSTRACT FROM AUTHOR]

Textual reviews contain fine-grained information that can effectively infer user preferences over the items. Accordingly, the latest studies in the field of recommender systems exploit content-rich review texts to complement user and item representations and improve the ability to make personalized recommendations. Furthermore, the interactive deep learning mechanism can better model the user-item interaction from fine-grained textual reviews compared to traditional recommendation approaches improving the predictive performances. Therefore, it becomes important to investigate the design of existing deep learning methods for review-based recommender systems and innovate to make them capable of meeting desired recommendation schemes. The purpose of this research is to explore the performance of deep learning networks and deep transformer models in review-based recommender systems. In this paper, we conduct a compendious survey of the latest deep learning techniques in review-based recommender systems. Then investigation calls to employ and analyze deep transformer models for the review-based recommender systems. The wide range of experiments shows that deep transformer models can extract interpretable and relevant user/item representations than traditional deep learning networks. The findings indicate that the best deep transformer performance gains the maximum relative improvement (RMSE = 4.6%, MAE = 7.4%) with Amazon electronics, compared to the best outcome from traditional deep learning networks. In the end, this article highlights research gaps and outlines research opportunities for future research in this field. [ABSTRACT FROM AUTHOR]

Deep learning architectures have demonstrated outstanding results in image classification in the last few years. However, applying sophisticated neural network architectures in small datasets remains challenging. In this context, transfer learning is a promising approach for dealing with this scenario. Generally, the available pre-trained architectures adopt a standard fixed input, which usually implies resizing and cropping the input images in the preprocessing phase, causing information loss. Besides, images present visual features in different scales in real-world scenarios, and most common approaches do not consider this fact. In this paper, we propose an approach that applies transfer learning for dealing with small datasets and leverages visual features extracted by pre-trained models from different scales. We based our approach on graph convolutional networks (GCN) that take graphs representing the images in different scales as input and whose nodes are characterized by features extracted by pre-trained models from regular image patches of different scales. Since GCN can deal with graphs with different numbers of nodes, our approach can deal naturally with images of heterogeneous sizes without discarding relevant information. We evaluated our approach in two datasets: a set of geological images and a publicly available dataset, both presenting characteristics that challenge traditional approaches. We tested our approach by adopting three different pre-trained models as feature extractors: two efficient pre-trained CNN models (DenseNet and ResNeXt) and one Vision Transformer (CLIP). We compared our approach with two conventional approaches for dealing with image classification. The experiments show that our approach achieves better results than the conventional approaches for this task. • We propose multiscale patch-based graphs to represent image information. • Transfer learning is adopted for dealing with small datasets. • The approach can handle images with heterogeneous sizes. • The approach represents visual information at multiple scales. • The approach is base on convolutional graph neural networks. [ABSTRACT FROM AUTHOR]

With the rapid development of deep learning, emotion recognition from facial expressions has also greatly improved, but there are still limitations in terms of reliability when applied to the real world. In other words, facial expressions and the corresponding true emotions may be inconsistent. So, although audio or bio signals are additionally used to improve the reliability of emotion recognition, reasonable estimation performance or real-time operation of emotion recognition is not guaranteed yet. This paper addresses both of the above challenging issues. Each sensor input and feature extraction process is totally and asynchronously processed through a parallel processing library. And then, we predict the comprehensive state of the subject's internal/external emotions through modality-adaptive fusion, which considers the influence of the features of each modality. We verified the performance of the proposed system through a real-time pilot test, and achieved accuracy of up to 33% higher than emotion recognition using only external signals, i.e., video and audio. • Real-time multimodal emotion recognition with visual, audio, and biological signals. • The mismatch between the inner and external expressions was considered first time. • An adaptive fusion method for adaptively analyzing a subject's internal/external signals. • Our system can capture true emotions of subjects in a real-world environment. [ABSTRACT FROM AUTHOR]

• This paper proposes a MOMHTS optimized HRFDL architecture for diagnosing the COVID-19 and pneumonia from chest CT and X-ray images via a collaborative edge framework to offer web services in underprivileged areas. • The hyperparameters of the HRFDL classifier are optimized using the MOMHTS algorithm which minimizes the complexity of the Deep learning model that prevents it from being deployed in the edge devices. COVID-19 is a global pandemic that mostly affects patients' respiratory systems, and the only way to protect oneself against the virus at present moment is to diagnose the illness, isolate the patient, and provide immunization. In the present situation, the testing used to predict COVID-19 is inefficient and results in more false positives. This difficulty can be solved by developing a remote medical decision support system that detects illness using CT scans or X-ray images with less manual interaction and is less prone to errors. The state-of-art techniques mainly used complex deep learning architectures which are not quite effective when deployed in resource-constrained edge devices. To overcome this problem, a Multi-Objective Modified Heat Transfer Search (MOMHTS) optimized Hybrid Random Forest Deep learning (HRFDL) classifier is proposed in this paper. The MOMHTS algorithm mainly optimizes the deep learning model in the HRFDL architecture by optimizing the hyperparameters associated with it to support the resource-constrained edge devices. To evaluate the efficiency of this technique, extensive experimentation is conducted on two real-time datasets namely the COVID19 lung CT scan dataset and the Chest X-ray images (Pneumonia) datasets. The proposed methodology mainly offers increased speed for communication between the IoT devices and COVID-19 detection via the MOMHTS optimized HRFDL classifier is modified to support the resources which can only support minimal computation and handle minimum storage. The proposed methodology offers an accuracy of 99% for both the COVID19 lung CT scan dataset and the Chest X-ray images (Pneumonia) datasets with minimal computational time, cost, and storage. Based on the simulation outcomes, we can conclude that the proposed methodology is an appropriate fit for edge computing detection to identify the COVID19 and pneumonia with higher detection accuracy. [ABSTRACT FROM AUTHOR]

The past decade has witnessed the rapid development of Artificial Intelligence (AI), especially the explosion of deep learning-related connectionist approaches. This study combines traditional literature review, bibliometric methods, and the Science of Science (SciSci) theory to scrutinize the development context of AI in the last decade on AMiner. 4 4 www.aminer.cn , AMiner is an academic mining system, which will be introduced in the following section. With the assistance of AMiner tools and datasets, this paper aims to describe a further explicit context and evolution of AI in the past decade from the development of connectionist approaches. Five aspects of the past decade are highlighted: self-learning and self-coding algorithms, Recurrent Neural Networks (RNN) algorithms, reinforcement learning, pre-trained models, and other typical deep learning algorithms, which represent the significant progress of this field. By combining these critical parts, we then summarize the current limitations and corresponding future of AI trends in the next decade and discuss some topics about the next generation of AI. Discoveries in this paper will benefit AI research in promoting understanding of the current critical stage and future trends of AI development and the AI industry in the dramatic ascendant for the academic research results transformation and its industrial layout. [ABSTRACT FROM AUTHOR]

With the rapid growth of the scale of social media information, it is getting more and more difficult for social media to detect fake news by using manual review. The spread of fake news may misguide the public, cause social panic, and even lead to violence, which could be avoided by using early detection technology to timely identify fake news on social media. Since fake news is often deliberately designed to attract attention, it is difficult for mongers to provide pictures that match the fabricated stories. However, most of existing multi-modal solutions only use the information of images and text, but do not take into account the correlation between them, which limits the effect of model detection effect. In this paper, we proposed a novel Fake News Detection Framework (FNDF) in Sina Weibo based on hybrid feature fusion method. Specifically, a total of 16 features from text, images and users are extracted to distinguish fake news. Moreover, we extract image-text correlation between text and images. Then, a new deep neural network model called Fake News Net (FNN) is built to implement the detection of fake news, which makes use of a pre-training model named Enhanced Representation through Knowledge Integration (ERNIE), a convolution network named Visual Geometry Group (VGG-19), and a Back Propagation (BP) neural network. We validated it on a publicly available dataset, which shows that the F1-score of the FNN model reaches 95.90%, outperforming the state-of-the-art methods by 3.08%. The ablation experiment also proves that the correlation between images and texts increased the F1-score of the model by 3.15%. And the data balancing experiments show that our model still keeps outstanding detection performance when there is less fake news compared to real news, which is closer to the real-world scenario. The research in this paper provides theoretical methods and research ideas for the detection of fake news on social networks. • A framework FNDF extracts a total of 16 features from text, images and users respectively. • The semantic features of text in images are deeply mined. • Based on SiameseNet, this paper proposes a feature of image-text correlation. • A novel deep neural network called FNN is built for hybrid feature fusion. • The results demonstrate that our method outperforms the state-of-the-art models. [ABSTRACT FROM AUTHOR]

Globally, all the major crops are significantly affected by diseases every year, as manual inspection across diverse fields is time-consuming, tedious, and requires expert knowledge. This leads to significant crop loss in different parts of the world. To provide effective solutions, several smart agriculture solutions are deployed for the control of pests and plant diseases using vision-based machine learning techniques. Despite rapid growth in the field, not many methods have been explored for their suitability in real-time applications. Several open challenges need to be addressed for the applicability of machine learning techniques in IoT-based smart agriculture solutions. Starting from data capturing methods and the availability of public datasets, the present paper provides a comprehensive review of vision-based machine learning techniques for plant disease detection. Initially, 1337 articles were selected from various scholarly resources to perform the survey. Based on the saliency of approaches, 148 articles are reviewed in this paper. Interestingly, a significant amount of research in this direction is taken up by Chinese and Indian researchers, and deep learning is the current research trend, as in other fields. The review concludes that a majority of existing methods exhibit their efficacy on public datasets captured mostly in controlled environmental conditions, but their generalization capability for in-field plant disease detection has not been explored. Lightweight CNN-based methods, on the other hand, have been designed for a limited number of diseases only, and are generally trained on small datasets. The scarcity of large-scale, in-field public datasets is one of the major bottlenecks in developing solutions that can work for a wide variety of plant diseases. [ABSTRACT FROM AUTHOR]

Vertical federated learning (VFL) is a privacy preserving collaborative machine learning technique designed for distributed learning scenarios in which data from different parties have overlap in the sample space. In this paper, a VFL method for feature selection, which is an effective dimensionality reduction technique that selects a subset of informative features from high-dimensional data by eliminating irrelevant and redundant features, is proposed. Because of the potential insufficiency of useful information for learning informative features and the difficulty in sharing raw data among parties due to the increasing awareness of data privacy protection, it is desirable to exploit information from multiple parties without raw data sharing. In this paper, we propose a VFL-based feature selection method that leverages deep learning models as well as complementary information from features in the same samples at multiple parties without data disclosure. In order to further improve feature selection performance, information of samples that do not have features appearing in all parties are also utilized. Promising results in extensive experiments show the effectiveness of the proposed approach in terms of collaborative feature selection without data sharing. • In this paper, we bridge this gap by proposing a novel VFL-based feature selection method—Vertical Federated Learning-based Feature Selection (VFLFS). To the best of our knowledge, this is the first deep learning-based vertical federated learning approach with feature selection. • A strategy to make use of non-overlapping samples is also proposed to improve feature selection effectiveness. • The proposed VFLFS approach has been evaluated extensively based on real-world datasets. The results show that VFLFS can significantly improve model performance under VFL settings compared to four state of the art baselines, especially in conditions where a large proportion of the data samples do not overlap across data owners. [ABSTRACT FROM AUTHOR]

With the rise in popularity of social media platforms and online forums, they have become a global source of news. Fake News (FNs) spreading through numerous institutions and sectors jeopardizes their reputations, causing users to abandon these platforms. Therefore, there is a huge pool of research in the area of Artificial Intelligence (AI) techniques that are used to detect FNs. Previously, great focus was given to online review classification and free internet posts based on social networks. This research proposes a Deep Learning-based FNs Detection method. This paper proposes a Deep Learning (DL)-based method for detecting FNs. The proposed system consists of three phases; text encoding, feature extraction, and classification. The text encoding process is carried out on the input news words using GLOVE for word representation. The encoded words are then embedded into a specific word length in order to be enrolled in the proposed DL models. The proposed DL models comprise both automatic feature extraction and classification tasks. Furthermore, this study proposes four different DL models, including Convolutional Neural Networks (CNNs) and Concatenated CNNs (C-CNNs), long short-term memory (LSTM), and Gated Recurrent Units, to find an optimal model prior to the section of FNs that outperforms previous works.The proposed DL models are carried out on FNs and FNC datasets which are provided by kaggle, and the suggested C-CNNs algorithm obtained an accuracy of 99.6% and trained faster than others. Multiple evaluation metrics such as precision, recall, F1, and accuracy have been utilized to evaluate the outcome of the proposed models. The experimental results demonstrated overall improvements in the subject of FND when compared with the current models and validated the potential of the proposed methodology for the detection of FNs on Social Media (SM). This study will help researchers to broaden the knowledge of applications of CNNs based on DL methods for FND. • This paper proposes a solution for fake news detection based on deep learning. • The proposed deep learning approach comprises both sequential and recurrent modalities. • The proposed sequential models includes CNNs and concatenated CNNs, while the recurrent includes LSTM and GRU models. • Each model is compared with the other proposed models as well as the works in the literature. [ABSTRACT FROM AUTHOR]

This paper presents a novel solution to automatically count vehicles in a parking lot using images captured by smart cameras. Unlike most of the literature on this task, which focuses on the analysis of single images, this paper proposes the use of multiple visual sources to monitor a wider parking area from different perspectives. The proposed multi-camera system is capable of automatically estimating the number of cars present in the entire parking lot directly on board the edge devices. It comprises an on-device deep learning-based detector that locates and counts the vehicles from the captured images and a decentralized geometric-based approach that can analyze the inter-camera shared areas and merge the data acquired by all the devices. We conducted the experimental evaluation on an extended version of the CNRPark-EXT dataset, a collection of images taken from the parking lot on the campus of the National Research Council (CNR) in Pisa, Italy. We show that our system is robust and takes advantage of the redundant information deriving from the different cameras, improving the overall performance without requiring any extra geometrical information of the monitored scene. • Smart mobility is crucial for smart cities and traffic-related issues. • We introduce a multi-camera system able to count cars from images of parking areas. • We combine a deep learning-based technique and a decentralized geometric approach. • All the algorithms run on the edge devices reducing the traffic on the network. • Our solution benefits from redundant information from different data sources. [ABSTRACT FROM AUTHOR]

Traffic forecasting is important for the success of intelligent transportation systems. Deep learning models, including convolution neural networks and recurrent neural networks, have been extensively applied in traffic forecasting problems to model spatial and temporal dependencies. In recent years, to model the graph structures in transportation systems as well as contextual information, graph neural networks have been introduced and have achieved state-of-the-art performance in a series of traffic forecasting problems. In this survey, we review the rapidly growing body of research using different graph neural networks, e.g. graph convolutional and graph attention networks, in various traffic forecasting problems, e.g. road traffic flow and speed forecasting, passenger flow forecasting in urban rail transit systems, and demand forecasting in ride-hailing platforms. We also present a comprehensive list of open data and source codes for each problem and identify future research directions. To the best of our knowledge, this paper is the first comprehensive survey that explores the application of graph neural networks for traffic forecasting problems. We have also created a public GitHub repository where the latest papers, open data, and source codes will be updated. • The latest application of graph neural network in traffic forecasting is presented. • The often-ignored implementation and reproducibility in other surveys are examined. • The future directions along with the research frontiers are pointed out. [ABSTRACT FROM AUTHOR]

• A comprehensive review for mobile robots Visual Slam application. • Including the state-of-the-art deep learning methods for Visual Slam. • Simulation results for each section based on feature extraction techniques. This paper is an overview to Visual Simultaneous Localization and Mapping (V-SLAM). We discuss the basic definitions in the SLAM and vision system fields and provide a review of the state-of-the-art methods utilized for mobile robot's vision and SLAM. This paper covers topics from the basic SLAM methods, vision sensors, machine vision algorithms for feature extraction and matching, Deep Learning (DL) methods and datasets for Visual Odometry (VO) and Loop Closure (LC) in V-SLAM applications. Several feature extraction and matching algorithms are simulated to show a better vision of feature-based techniques. [ABSTRACT FROM AUTHOR]

• Health indicators construction techniques of bearing and gear are reviewed. • Methods of construction, pros and cons of each technique are addressed. • Current challenges and research prospects on health indicator construction are given. The condition monitoring (CM) of rotating machinery (RM) is an essential operation for improving the reliability of mechanical systems. For this purpose, an efficient CM method that possesses simple and intuitive attributes is required for industrial applications. For condition monitoring that connects fault detection, degradation assessment, and prognosis applications, health indicators (HIs) have been developed in the past few decades. The construction of a HI is the decisive procedure for extracting informative fault information from the monitoring signal. From the initial statistical parameter-based construction methods to the introduction of data-oriented intelligent methods such as deep learning in recent years, HIs construction methods have ranged from fault mechanism-based approaches to a data-based approach, which involve two different technologies regardless of superiority or inferiority. This paper provides a systematic review of the HIs construction methods for rotating machinery proposed in the literature. It emphasizes the classical technical approaches and recent interesting research trends and analyzes the benefits and potential of efficient HIs for condition monitoring. The current challenges and future research opportunities are also presented in this paper. The Engineers and researchers interested in this research can be informed of current research ideas and directions in the field by reading this paper, as well as inspiring potentially excellent research work in the future. [ABSTRACT FROM AUTHOR]

Drug design is a research process with a goal of creating a chemical drug to produce the desired biological effect. Because of the long time and the high cost issues associated with traditional drug discovery, there is a need to develop new techniques and strategies to increase the diminishing effectiveness of traditional approaches. Ligand-Based Virtual Screening (LBVS) plays a vital role in the early stage of the drug discovery. It could constitute a possible solution to solve the time and cost problems. Subsequently, the researchers are looking for new methods to find new active compounds and bring them to market in a short time. LBVS can be enhanced by different methods and strategies such as Machine Learning and Deep Learning. In this paper, a Deep Convolutional Neural Network method is proposed to enhance the performances of Ligand-Based Virtual Screening process (DCNNLB). Two main contributions are presented in this paper, The first contribution consists of designing a model based on Deep Convolutional Neural Network (DCNN) for LBVS. We propose several topological network models to find the one that gives the best performance such as accuracy and recall. For this, many network topology configurations have been proposed, and a variety of parameters have been taken into account. Furthermore, our proposed model is trained on all compounds of all activity classes of the Drug Data Report Database (MDDR). Thus, it presented a mean accuracy of 0.98 for all three MDDR Datasets. The second contribution is to generate a new learning representation in order to better represent chemical compound. This representation is based on the extraction of the automatic features learning from the weights of our proposed model. Consequently, it is very efficient in calculating molecular similarity and performances of the LBVS process. The obtained results with the three different datasets drawn from the MDDR and the performance evaluation with ANOVA test, have proved the superiority in performance of our proposed method compared to the different conventional methods. [Display omitted] • New molecular representations obtained by the proposed deep convolutional model. • Enhancement in effectiveness of similarity searching with DCNNLB method in LBVS. • Decreasing in training time by using the molecular learning representation. • The proposed DCNNLB method significantly outperforms the conventional TAN method. [ABSTRACT FROM AUTHOR]

Financial markets play an important role on the economical and social organization of modern society. In these kinds of markets, information is an invaluable asset. However, with the modernization of the financial transactions and the information systems, the large amount of information available for a trader can make prohibitive the analysis of a financial asset. In the last decades, many researchers have attempted to develop computational intelligent methods and algorithms to support the decision-making in different financial market segments. In the literature, there is a huge number of scientific papers that investigate the use of computational intelligence techniques to solve financial market problems. However, only few studies have focused on review the literature of this topic. Most of the existing review articles have a limited scope, either by focusing on a specific financial market application or by focusing on a family of machine learning algorithms. This paper presents a review of the application of several computational intelligent methods in several financial applications. This paper gives an overview of the most important primary studies published from 2009 to 2015, which cover techniques for preprocessing and clustering of financial data, for forecasting future market movements, for mining financial text information, among others. The main contributions of this paper are: (i) a comprehensive review of the literature of this field, (ii) the definition of a systematic procedure for guiding the task of building an intelligent trading system and (iii) a discussion about the main challenges and open problems in this scientific field. [ABSTRACT FROM AUTHOR]

• A embedding method MAHE-IM for IM on heterogeneous network. • MAHE-IM extracts and aggregates multiple heterogeneous relationships. • Eight network embedding and seven GNN methods extended for IM. • MAHE-IM outperforms eighteen state-of-the-art algorithms. • A web server (https://mahe-im.com/) is developed for maximizing user convenience. Influence maximization (IM), as an essential problem in social network analysis, can identify a minimum group of the influential nodes to maximize the spread of information on the network. The majority of IM studies focus on homogeneous networks, whereas in real world, heterogeneous networks are ubiquitous. The existing IM methods based on homogeneous networks do not consider a variety of complex heterogeneous relationships and the attribute of different types of nodes, which don't accommodate most real scenario. It is of great significance to study IM methods based on heterogeneous networks, where efficient integrating complex multiple semantic relationships and structures embedded in the heterogeneous network is the key breakthrough point. In this paper, a novel deep learning algorithm for influence maximization on heterogeneous networks based on a multiple aggregation of heterogeneous relation embedding named MAHE-IM is proposed, which can capture the heterogeneous high-order structure and semantic features of heterogeneous information networks. For more comprehensive and systematical evaluation of MAHE-IM, we further extend fourteen state-of-the-art homogeneous and heterogeneous network embedding and graph neural network methods for IM problem and propose fourteen extented IM algorithms. Four popular IM algorithms and our extended fourteen IM algorithms are taken as eighteen baseline algorithms, which can be categorized three types: greedy-based, network embedding-based and GNN-based IM algorithms. Compared with eighteen baseline algorithms, the experimental results illustrate that MAHE-IM is significantly efficient and effective. MAHE-IM is more practical on heterogeneous networks. In addition, in order to maximize the convenience for users, a webserver (https://mahe-im.com/) is developed. The users can obtain the IM results by submitting their heterogeneous networks to the webserver. The corresponding source code and used data in this paper can also be available at https://mahe-im.com/ and https://codeocean.com/capsule/4091031/tree/v1. [ABSTRACT FROM AUTHOR]

• An automated system for railgun launcher damage detection is proposed. • A new data set of railgun inner bore damage is obtained by the designed device. • Adaptive data augmentation and focal loss are used to balance the uneven data set. • YOLOv5 and SOLOv2 are used for the detection and shape extraction of damage. • Results on the data set show the effectiveness of the proposed methods. The inner bore damage affects the launch performance and service life of electromagnetic railgun launcher. Detection and observation of railgun inner bore damage contribute to the study on mechanism and development rules of railgun damage. This paper analyzes five types of typical railgun inner bore damage. Based on the detection requirement for the damages, this paper proposes an automated damage detection system for the inner bore of electromagnetic railgun launcher consisting of data acquisition device and detection algorithms. The proposed device can step inside the inner bore of the railgun launcher to take photos of the inner bore surface automatically. We use the images obtained by the proposed device to build a data set for the training and verification of the detection algorithms. The object detection algorithm You Only Look Once v5 (YOLOv5) is utilized to achieve the rapid detection of railgun inner bore damages. We introduce the adaptive data augmentation and the focal loss to balance out the uneven category distribution of our data set. The result proves that our YOLOv5 model reaches the state-of-the-art level in the railgun inner bore damage detection task, with its mean Average Precision (mAP) of 0.659 and detection speed of 47.6 frames per second (fps). We choose Segmenting Objects by Locations v2 (SOLOv2) to extract the shape of the damage, with the Average Precision of 0.631. We further achieve damage statistics and the model visualization of damage distribution. The experimental results show that the proposed detection system meets the requirements of rapid detection and accurate feature extraction. It provides researchers with an approach for the study of railgun inner bore damage mechanism. [ABSTRACT FROM AUTHOR]

Prediction of clinical outcomes using the patient's medical data enhances clinical decision making and improves prognostic accuracy. Deep learning (DL) for medical decision support systems has particularly shown expert-level accuracy in predicting clinical outcomes. However, most of these machine learning and Artificial Intelligent (AI) models lack interpretability which causes major trust-related problems in healthcare. This necessitates the need for interpretable AI systems that can explain their decisions. This paper is designed to address the problem of how clinical decision support systems can be designed to be transparent, interpretable and therefore comprehensible for humans. In this view, this paper proposes an attentive hierarchical adaptive neuro-fuzzy inference system (AH-ANFIS), that combines fuzzy inference in a hierarchical architecture with attention for selecting the important rules. The proposed model benefits from the rule-based structure of ANFIS that enables the user to interpret the abstractions of hidden layers. The hierarchical structure in fuzzy modeling helps to overcome the rule explosion problem that arises with large medical data and improve interpretability. To ensure the improvement of interpretability with hierarchical architecture, an evolutionary algorithm (EA) is used to decompose the input space into an optimal permutation of input subsets which results in subsystems with independent meaning. The attention-based rule selector identifies the most activated rule to select important patient-specific features for predicting the clinical outcome. To verify the performance of the proposed AH-ANFIS and analyze the important features for classification, we perform experiments with two cancer diagnostic datasets. By pruning fuzzy if-then rules using recursive rule elimination (RRE), the complexity of the model is largely reduced while maintaining the performance of the system making it more interpretable. • The AH-ANFIS combines fuzzy inference in a hierarchical architecture. • The hierarchical structure in fuzzy modeling overcome the rule explosion problem. • Interpret the system's conclusions based on fuzzy rules extraction. • An EA is used to decompose the input space into the optimal input subsets. • Improve the interpretability by pruning fuzzy if-then rules. [ABSTRACT FROM AUTHOR]

Abnormal event detection in video is alternatively known as outlier detection, where machine learning can be highly effective. While testing an unknown video, the objective of such methods is to verify the video's category, e.g. normal or abnormal. This paper exploits visual information from normal as well as abnormal videos to train a deep multiple instance learning classifier for classification of videos. Existing multiple instance learning classifiers presume that the training videos contain only short-duration anomalous events. This assumption may not be valid for all real-world anomalies. Also, multiple occurrences of anomalies in training videos cannot be ruled out. This paper shows that by injecting temporal information in feature extraction, anomaly detection performance can be improved. To accomplish this, two spatio-temporal deep feature extractors have been applied in parallel on the training videos. These streams are then used to train a modified multiple instance learning-based classifier. Finally, a fuzzy aggregation is employed to fuse the anomaly scores. Additionally, two lightweight deep-learning classifiers have been used to substantiate the model's efficacy for classifying fire and accident events. To understand the reliability and performance of the proposed method, extensive experiments have been carried out using UCF-Crime video dataset containing 13 anomaly categories. The dataset has been restructured into five broad categories based on the severity of actions to study the robustness of the proposed method. The paper provides sufficient empirical evidence which proves that by incorporating temporal features in the pipeline, anomaly detection accuracy can be significantly improved. Moreover, the model helps to detect long-duration anomalies in videos, which was not possible using existing methods. The proposed end-to-end multi-stream architecture performs abnormal event detection with accuracy as high as 84.48%, which is better than the performance of existing video anomaly detection methods. Moreover, the class-wise detection accuracy has improved by 6%–14% across various broad categories. • Multi-stream deep learning architecture for anomaly localization in videos. • Improvement in the conventional MIL classifier for binary classification. • A late fuzzy fusion technique to improve the anomaly localization scores. • Classification of road accident and fire related anomalies. [ABSTRACT FROM AUTHOR]

The ability of transformers to perform precision tasks such as question answering, Natural Language Inference (NLI) or summarizing, has enabled them to be ranked as one of the best paradigms to address Natural Language Processing (NLP) tasks. NLI is one of the best scenarios to test these architectures, due to the knowledge required to understand complex sentences and established relationships between a hypothesis and a premise. Nevertheless, these models suffer from the incapacity to generalize to other domains or from difficulties to face multilingual and interlingual scenarios. The leading pathway in the literature to address these issues involve designing and training extremely large architectures, but this causes unpredictable behaviors and establishes barriers which impede broad access and fine tuning. In this paper, we propose a new architecture called Siamese Inter-Lingual Transformer (SILT). This architecture is able to efficiently align multilingual embeddings for Natural Language Inference, allowing for unmatched language pairs to be processed. SILT leverages siamese pre-trained multi-lingual transformers with frozen weights where the two input sentences attend to each other to later be combined through a matrix alignment method. The experimental results carried out in this paper evidence that SILT allows to reduce drastically the number of trainable parameters while allowing for inter-lingual NLI and achieving state-of-the-art performance on common benchmarks. • Efficient alignment of multilingual embeddings for Natural Language Inference. • Siamese pretrained multilingual transformers with frozen weights and mutual attention. • A curated Spanish version of the SICK dataset, called SICK-es is provided. • Drastic reduction of trainable parameters and ability to perform inter-lingual tasks. [ABSTRACT FROM AUTHOR]

Human gaze estimation plays a major role in many applications in human–computer interaction and computer vision by identifying the users' point-of-interest. Revolutionary developments of deep learning have captured significant attention in gaze estimation literature. Gaze estimation techniques have progressed from single-user constrained environments to multi-user unconstrained environments with the applicability of deep learning techniques in complex unconstrained environments with extensive variations. This paper presents a comprehensive survey of the single-user and multi-user gaze estimation approaches with deep learning. State-of-the-art approaches are analyzed based on deep learning model architectures, coordinate systems, environmental constraints, datasets and performance evaluation metrics. A key outcome from this survey realizes the limitations, challenges and future directions of multi-user gaze estimation techniques. Furthermore, this paper serves as a reference point and a guideline for future multi-user gaze estimation research. • Presents an analysis of gaze estimation approaches in unconstrained environments. • Survey on existing gaze datasets and requirements for a multi-user gaze dataset. • Analyzes gaze estimation coordinate systems and performance evaluation metrics. • Analyzes challenges and future directions in gaze estimation. [ABSTRACT FROM AUTHOR]

Deep learning for surface defect detection has recently received increased attention in relation to quality control of industrial products. In this paper, we present a combined loss function called Segmentation–Classification combined Loss (SCL). It exploits the binary information (normal/anomalous sample) implicit in the segmentation masks to reduce false positives and false negatives in surface defect detection tasks. We evaluate the method on images in three different domains: defects in electrical commutators, defects in a weakly labeled synthetic dataset, and defects of glass bottles. Our combined loss function outperforms the state of the art in the Kolektor (electrical commutators) dataset and significantly improves the classical Dice segmentation loss function in terms of both average precision, false positives, and false negatives. Moreover, the combined loss function introduced in this paper can be applied in a straightforward manner to any loss-based segmentation model. Finally, in this work we introduce and publish a new dataset for the still poorly explored task of surface defect detection of glass bottles. • New multi-task combined loss for surface defect detection tasks. • New Oliena dataset for the poorly researched task of glass bottles defect detection. • Comparison between classification and segmentation networks for defect detection. • Comparison between a multi-task loss function and a multi-branch architecture. [ABSTRACT FROM AUTHOR]

• Simple first approach based on unsupervised Latent Semantic Indexing lead to state of the art results. • Deep learning approaches help avoid the time consuming step of feature engineering. • Pairwise approaches outperfom pointwise approaches. • Using overlapping terms as inputs to neural networks improve both results and performance. • Using pre-trained word embeddings reduce the overfitting of deep learning models. In this paper we face the problem of Community Question Answering for the Arabic language. In this setting, a member of the community posts an initial query expressed in Natural Language. Other participants post their own interventions: answers, comments, additional questions, etc. contributing to building a rather tangled thread of nodes containing Natural Language short texts. The task consists in answering the initial query using the thread as the space of possible answers. The task can be approached as a classification, a regression or a ranking problem. In our case we select the set of possible candidates, we assign a relevance score to each candidate and we rank them accordingly. We propose a bunch of unsupervised models and show that a model based on Latent Semantic Indexing approach outperforms state of the art models for this task. We also use transfer learning to power the embeddings layer of various deep learning models and prove that the pairwise approaches outperform their pointwise counterparts. All the proposed models have been evaluated on Semeval 2017 Task 3 Subtask D: Arabic Community Question Answering and achieve state of the art or near performance. [ABSTRACT FROM AUTHOR]

• A model based on deep learning for environmental sound classification is proposed. • Model provides the trade-off between the audio length, accuracy and parameter space. • Model omits the signal processing modules and learns representations from audio. • Model has a small parameter space compared to other architectures in the literature. • The frequency and magnitude responses of some of the learned filters are analyzed. In this paper, we present an end-to-end approach for environmental sound classification based on a 1D Convolution Neural Network (CNN) that learns a representation directly from the audio signal. Several convolutional layers are used to capture the signal's fine time structure and learn diverse filters that are relevant to the classification task. The proposed approach can deal with audio signals of any length as it splits the signal into overlapped frames using a sliding window. Different architectures considering several input sizes are evaluated, including the initialization of the first convolutional layer with a Gammatone filterbank that models the human auditory filter response in the cochlea. The performance of the proposed end-to-end approach in classifying environmental sounds was assessed on the UrbanSound8k dataset and the experimental results have shown that it achieves 89% of mean accuracy. Therefore, the proposed approach outperforms most of the state-of-the-art approaches that use handcrafted features or 2D representations as input. Moreover, the proposed approach outperforms all approaches that use raw audio signal as input to the classifier. Furthermore, the proposed approach has a small number of parameters compared to other architectures found in the literature, which reduces the amount of data required for training. [ABSTRACT FROM AUTHOR]