Search

After 70 years of intricate development, machine learning, represented by deep learning, is based on the multilevel structure of the human brain and the layer-by-layer analysis and processing mechanism of neuron connection and interaction information. The powerful parallel information processing ability of self-adaptation and self-learning has allowed for breakthroughs in many fields, among which the most representative is image recognition. Therefore, this paper proposed optimizing the RBF algorithm with machine learning (ML) to improve the recognition rate of spectrum sensing. The results showed that the average detection success rates of the RBF algorithm were 93.62%, 95.07%, 96.91%, 98.78% and 99.37% when the SNRs were − 8 dB, − 4 dB, 0 dB, 4 dB and 8 dB, respectively, and the other conditions were kept the same. The average detection success rates of the SVM/RBF algorithm were 97.65%, 99.63%, 99.76%, 99.91% and 99.88%, respectively. The average detection success rate of the SVM/RBF algorithm was significantly higher than that of the RBF algorithm. This indicates that analyzing the RBF neural network algorithm through ML can improve the success rate of spectrum sensing, which highlights a new direction for the application of ML and neural networks. [ABSTRACT FROM AUTHOR]

With the leaps and bounds of computer performance and the advent of the era of big data, deep learning has drawn more and more attention from all walks of life. It can combine low-level features to form more abstract high-level features and describe the data more essentially. Therefore, it is widely used in various fields such as computer vision. Image segmentation is one of the most basic research topics in the field of computer vision. The main purpose is to extract regions of interest from images for later image processing research. In 3D reconstruction based on sequence images, the segmentation accuracy and speed of sequence images determine the quality and efficiency of target reconstruction. Therefore, when facing large-scale sequence images, the biggest problem is how to improve the segmentation speed while ensuring accuracy. Based on the above background, the research content of this article is an indoor scene segmentation algorithm based on full convolutional neural network. According to the characteristics of indoor application scenes, this paper proposes a fast convolutional neural network image segmentation method to segment the indoor scene image and construct the fast fully convolutional networks (FFCN) for indoor scene image segmentation uses inter-layer fusion to reduce the amount of network calculation parameters and avoid the loss of picture feature information by continuous convolution. In order to verify the effectiveness of the network, in this paper, a basic living object data set (XAUT data set) in an indoor environment is created. The XAUT data set is used to train the FFCN network under the Caffe framework to obtain an indoor scene segmentation model. In order to compare the effectiveness of the model, the structure of the worn FCN8s, FCN16s, and FCN32s models was fine-tuned, and the corresponding algorithm model for indoor scene segmentation was obtained by training with the XAUT data set. The experimental results show that the pixel recognition accuracy of all types of networks has reached 86%, and the mean IU ratio has reached more than 63%. The mean IU of the FCN8s network is the highest at 70.38%, but its segmentation speed is only 1/5 of FFCN. On the premise that other types of indicators are not much different, the average segmentation speed on FFCN fast segmentation convolutional neural network reaches 40 fps. It can be seen that the scale fusion technology can well avoid the loss of image feature information in the network convolution and reddening process. Compared with other FCN networks, it has a faster speed and is conducive to real-time image preprocessing. [ABSTRACT FROM AUTHOR]

Detection of small objects is one of the main challenges to be improved in deep learning, mainly due to the small number of pixels and scene's context, leading to a loss in performance. In this paper, we present an optimized approach based on deep object detection models that allow the detection of a higher number of elements and improve the score obtained for their class inference. The main advantage of the presented methodology is that it is not necessary to modify the internal structure of the selected convolutional neural network model or re-training for a specific scene. Our proposal is based on detecting initial regions to generate several sub-images using super-resolution (SR) techniques, increasing the number of pixels of the elements, and re-infer over these areas using the same pre-trained model. A reduced set of windows is calculated in the super-resolved image by analyzing a computed graph that describes the distances among the preliminary object detections. This analysis is done by finding maximal cliques on it. This way, the number of windows to be examined is diminished, significantly speeding up the detection process. This framework has been successfully tested on real traffic sequences obtained from the U.S. Department of Transportation. An increase of up to 44.6% is achieved, going from an average detection rate for the EfficientDet D4 model of 14.5% compared to 59.1% using the methodology presented for the first sequence. Qualitative experiments have also been performed over the Cityscapes and VisDrone datasets. [ABSTRACT FROM AUTHOR]

Traditional navigation systems rely mainly on satellite navigation, but internal positioning and navigation cannot be achieved. This is mainly due to the inability to penetrate the wall due to the complex internal environment and the signal reaching the ground. Therefore, this paper proposes the research of indoor positioning algorithm based on improved convolutional neural network. The main method of this paper is based on the monocular vision indoor positioning based on the improved convolutional neural network. This method is an intelligent solution to the problems of traditional methods such as high cost, poor anti-interference ability, weak robustness, and poor compatibility. Positioning aids. The traditional computer vision camera pose estimation method is affected by the complex background in the image. When extracting feature corners, it is very affected by non-interest corners, so this paper adds an improved convolutional neural network algorithm to add in complex indoor scenes. With the region limitation, camera pose estimation in the region of interest is better to achieve low-cost, high accuracy and more stability. The improved model in this paper was tested on a test set of indoor datasets with markers, and the recognition accuracy reached 98.1%. In addition, this paper improves the PnP solution method, compares and analyzes the traditional RPnP, EPnP, and CEPPnP algorithms, and seeks the optimal camera pose estimation algorithm to locate the world coordinates of the current shooting position. The experimental results show that this paper proposes the positioning algorithm has high stability. [ABSTRACT FROM AUTHOR]

One of the most popular techniques for changing the purpose of an image or resizing a digital image with content awareness is the seam-carving method. The performance of image resizing algorithms based on seam machining shows that these algorithms are highly dependent on the extraction of importance map techniques and the detection of salient objects. So far, various algorithms have been proposed to extract the importance map. In this paper, a new method based on Rényi entropy is proposed to extract the importance map. Also, a deep learning network has been used to detect salient objects. The simulator results showed that combining Rényi's importance map with a deep network of salient object detection performed better than classical seam-carving and other extended seam-carving algorithms based on deep learning. [ABSTRACT FROM AUTHOR]

In sports, advance sensing technologies generate massive amount of unstructured telemedicine data that need to be refined for accurate diagnosis of underlying diseases. For accurate prediction of diseases and classification of athletes' data, deep learning algorithms are frequently used at the cloud. However, the transmission of raw data of athletes to the cloud faces numerous challenges. Among them, security and privacy are a major challenge in view of the sensitive and personal information present within the unstructured data. In this paper, first we present a data block scrambling algorithm (without key management) for secured transmission and storage of ECG (electrocardiogram) data of table tennis players at the cloud. A small piece of original data stored at the cloud is used for scrambling the massive amount of remaining ECG data. The secured telemedicine data is then imported into Hadoop Distributed File System for data management, which is read by Spark framework to form Resilient Distributed Datasets. Finally, a deep learning approach is used that extracts useful features, learns the related information, and weights and sums the feature vectors at different layers for classification. Theoretical analysis proves that our proposed approach is highly robust and resilient to brute force attacks and at the same time has a much better accuracy, sensitivity, and specificity as compared to the existing approaches. [ABSTRACT FROM AUTHOR]

This paper presents a digital image dataset of historical handwritten birth records stored in the archives of several parishes across Sweden, together with the corresponding metadata that supports the evaluation of document analysis algorithms' performance. The dataset is called SHIBR (the Swedish Historical Birth Records). The contribution of this paper is twofold. First, we believe it is the first and the largest Swedish dataset of its kind provided as open access (15,000 high-resolution colour images of the era between 1800 and 1840). We also perform some data mining of the dataset to uncover some statistics and facts that might be of interest and use to genealogists. Second, we provide a comprehensive survey of contemporary datasets in the field that are open to the public along with a compact review of word spotting techniques. The word transcription file contains 17 columns of information pertaining to each image (e.g., child's first name, birth date, date of baptism, father's first/last name, mother's first/last name, death records, town, job title of the father/mother, etc.). Moreover, we evaluate some deep learning models, pre-trained on two other renowned datasets, for word spotting in SHIBR. However, our dataset proved challenging due to the unique handwriting style. Therefore, the dataset could also be used for competitions dedicated to a large set of document analysis problems, including word spotting. [ABSTRACT FROM AUTHOR]

Deep neural networks (DNNs) have evolved as a beneficial machine learning method that has been successfully used in various applications. Currently, DNN is a superior technique of extracting information from massive sets of data in a self-organized method. DNNs have different structures and parameters, which are usually produced for particular applications. Nevertheless, the training procedures of DNNs can be protracted depending on the given application and the size of the training set. Further, determining the most precise and practical structure of a deep learning method in a reasonable time is a possible problem related to this procedure. Meta-heuristics techniques, such as swarm intelligence (SI) and evolutionary computing (EC), represent optimization frames with specific theories and objective functions. These methods are adjustable and have been demonstrated their effectiveness in various applications; hence, they can optimize the DNNs models. This paper presents a comprehensive survey of the recent optimization methods (i.e., SI and EC) employed to enhance DNNs performance on various tasks. This paper also analyzes the importance of optimization methods in generating the optimal hyper-parameters and structures of DNNs in taking into consideration massive-scale data. Finally, several potential directions that still need improvements and open problems in evolutionary DNNs are identified. [ABSTRACT FROM AUTHOR]

The automatic detection of software vulnerabilities is considered a complex and common research problem. It is possible to detect several security vulnerabilities using static analysis (SA) tools, but comparatively high false-positive rates are observed in this case. Existing solutions to this problem depend on human experts to identify functionality, and as a result, several vulnerabilities are often overlooked. This paper introduces a novel approach for effectively and reliably finding vulnerabilities in open-source software programs. In this paper, we are motivated to examine the potential of the clonal selection theory. A novel deep learning-based vulnerability detection model is proposed to define features using the clustering theory of the clonal selection algorithm. To our knowledge, this is the first time we have used deep-learned long-lived team-hacker features to process memories of sequential features and mapping from the entire history of previous inputs to target vectors in theory. With an immune-based feature selection model, the proposed approach aimed to improve static analyses' detection abilities. A real-world SA dataset is used based on three open-source PHP applications. Comparisons are conducted based on using a classification model for all features to measure the proposed feature selection methods' classification improvement. The results demonstrated that the proposed method got significant enhancements, which occurred in the classification accuracy also in the true positive rate. [ABSTRACT FROM AUTHOR]

The goal of steganalysis is to unearth a concealed message hidden inside an innocent carrier by steganography. Steganography in the hands of unlawful people can pose a great threat to society. Secret message is embedded as a noise (residue) which silently disrupts the statistical nature of the carrier without visual transformation. For the purpose of unearthing covert communication, modeling this noise (residue) is a crucial factor. In this paper, a hybrid deep learning framework is proposed with convolutional neural network (CNN) to emphasize the need for residual based investigation for digital image steganalysis. In order to model the noise residuals, five types of handcrafted features are extracted, in which the first four types rely on acquiring noise residuals based on filtering by linear and nonlinear filters, and the fifth type relies on residual acquisition method based on Empirical Mode Decomposition (EMD). The extracted features are categorized using a robust CNN with a new parallel architecture that is capable of learning intricate details from input features to classify it as cover or stego. Mostly CNNs are trained with raw images, but in the proposed method, the 1-dimensional residual features are stacked into 2-dimensional grid and are then used to train the CNN. Three spatial content-adaptive and four spatial non-content-adaptive algorithms are used to evaluate the performance of the proposed architecture. The experimental results show the versatility and robustness of the proposed hybrid architecture towards these algorithms when compared to existing state-of-the-art steganalytic methods. A practical issue prevalent in deploying steganalysis in real-world is 'mismatch scenario,' and based on the experimentation done in this paper, the proposed architecture performs well under stego algorithm mismatch and payload mismatch scenarios. [ABSTRACT FROM AUTHOR]

Multiobjective evolutionary algorithms (MOEAs) have received much attention in multiobjective optimization in recent years due to their practicality. With limited computational resources, most existing MOEAs cannot efficiently solve large-scale multiobjective optimization problems (LSMOPs) that widely exist in the real world. This paper innovatively proposes a dual decomposition strategy (DDS) that can be embedded into many existing MOEAs to improve their performance in solving LSMOPs. Firstly, the outer decomposition uses a sliding window to divide large-scale decision variables into overlapped subsets of small-scale ones. A small-scale multiobjective optimization problem (MOP) is generated every time the sliding window slides. Then, once a small-scale MOP is generated, the inner decomposition immediately creates a set of global direction vectors to transform it into a set of single-objective optimization problems (SOPs). At last, all SOPs are optimized by adopting a block coordinate descent strategy, ensuring the solution's integrity and improving the algorithm's performance to some extent. Comparative experiments on benchmark test problems with seven state-of-the-art evolutionary algorithms and a deep learning-based algorithm framework have shown the remarkable efficiency and solution quality of the proposed DDS. Meanwhile, experiments on two real-world problems show that DDS can achieve the best performance beyond at least one order of magnitude with up to 3072 decision variables. [ABSTRACT FROM AUTHOR]

As the main component of the new energy battery, the safety vent usually is welded on the battery plate, which can prevent unpredictable explosion accidents caused by the increasing internal pressure of the battery. The welding quality of safety vent directly affects the safety and stability of the battery; so, the welding-defect detection is of great significance. In this paper, we researched the welding-defect detection method based on semantic segmentation algorithm. The automatic detection method should recognize, locate, and count the area of defects. However, small differences between inter-classes, strong variations in defect size, and uncertain annotation create challenges for defect semantic segmentation. To address these challenges, a customized deep learning model based on two-branch architecture is proposed to segment welding-defect at the pixel level. This architecture involves three modules as follows, (i) The Spatial Branch is elaborately designed to capture spatial details and generate high-resolution feature representation; (ii) The Context Branch is responsible for obtaining semantic context; (iii) The Feature Fusion Block is proposed to merge the two types of feature representation and enhance mutual connections. Furthermore, a welding image dataset was built to validate our method, and the network's performance of defect segmentation and classification were evaluated by the defects images and the normal images, respectively. Data augmentation strategies are performed to overcome the problem of imbalanced distribution and inaccurate labels of the dataset, which is caused by the ambiguity of defect edges and the subjectivity of manual annotation. The experiment results indicate that our method achieves 86.704 % of mIoU (mean intersection of union), which supports the effectiveness of the deep learning model in segmenting defects. Besides, 98.896 % of AP (average precision) and 0 % of MDR (miss detection rate) further suggests the applicability of the proposed framework in industrial applications. [ABSTRACT FROM AUTHOR]

Breast cancer is the second leading cause of death in women; therefore, effective early detection of this cancer can reduce its mortality rate. Breast cancer detection and classification in the early phases of development may allow for optimal therapy. Convolutional neural networks (CNNs) have enhanced tumor detection and classification efficiency in medical imaging compared to traditional approaches. This paper proposes a novel classification model for breast cancer diagnosis based on a hybridized CNN and an improved optimization algorithm, along with transfer learning, to help radiologists detect abnormalities efficiently. The marine predators algorithm (MPA) is the optimization algorithm we used, and we improve it using the opposition-based learning strategy to cope with the implied weaknesses of the original MPA. The improved marine predators algorithm (IMPA) is used to find the best values for the hyperparameters of the CNN architecture. The proposed method uses a pretrained CNN model called ResNet50 (residual network). This model is hybridized with the IMPA algorithm, resulting in an architecture called IMPA-ResNet50. Our evaluation is performed on two mammographic datasets, the mammographic image analysis society (MIAS) and curated breast imaging subset of DDSM (CBIS-DDSM) datasets. The proposed model was compared with other state-of-the-art approaches. The obtained results showed that the proposed model outperforms the compared state-of-the-art approaches, which are beneficial to classification performance, achieving 98.32% accuracy, 98.56% sensitivity, and 98.68% specificity on the CBIS-DDSM dataset and 98.88% accuracy, 97.61% sensitivity, and 98.40% specificity on the MIAS dataset. To evaluate the performance of IMPA in finding the optimal values for the hyperparameters of ResNet50 architecture, it compared to four other optimization algorithms including gravitational search algorithm (GSA), Harris hawks optimization (HHO), whale optimization algorithm (WOA), and the original MPA algorithm. The counterparts algorithms are also hybrid with the ResNet50 architecture produce models named GSA-ResNet50, HHO-ResNet50, WOA-ResNet50, and MPA-ResNet50, respectively. The results indicated that the proposed IMPA-ResNet50 is achieved a better performance than other counterparts. [ABSTRACT FROM AUTHOR]

Label manipulation attacks are a subclass of data poisoning attacks in adversarial machine learning used against different applications, such as malware detection. These types of attacks represent a serious threat to detection systems in environments having high noise rate or uncertainty, such as complex networks and Internet of Thing (IoT). Recent work in the literature has suggested using the K-nearest neighboring algorithm to defend against such attacks. However, such an approach can suffer from low to miss-classification rate accuracy. In this paper, we design an architecture to tackle the Android malware detection problem in IoT systems. We develop an attack mechanism based on silhouette clustering method, modified for mobile Android platforms. We proposed two convolutional neural network-type deep learning algorithms against this Silhouette Clustering-based Label Flipping Attack. We show the effectiveness of these two defense algorithms—label-based semi-supervised defense and clustering-based semi-supervised defense—in correcting labels being attacked. We evaluate the performance of the proposed algorithms by varying the various machine learning parameters on three Android datasets: Drebin, Contagio, and Genome and three types of features: API, intent, and permission. Our evaluation shows that using random forest feature selection and varying ratios of features can result in an improvement of up to 19% accuracy when compared with the state-of-the-art method in the literature. [ABSTRACT FROM AUTHOR]

E-commerce platforms apply recommendation technology to a wide variety of commercial websites to help users quickly find the product they need from a large number of products. At present, recommendation systems have been widely used in large-scale e-commerce websites, and most e-commerce shopping websites attract customers through the image information of goods. In this paper, the research status of content-based image retrieval algorithms is analysed, and how to use image content information to recommend goods is studied. A hierarchical commodity classification and retrieval system are designed. A class decision layer is used to determine the category of commodity images and then precisely retrieve the corresponding category of commodity image features. The corresponding relationship between different commodity visual features is used to recommend commodities, and a matching recommendation algorithm for commodities is proposed. Experiments show that the proposed image content-based recommendation method can coordinate with features, and its recommendation results have high accuracy in practical applications. [ABSTRACT FROM AUTHOR]

Univariate time series forecasting is still an important but challenging task. Considering the wide application of temporal data, adaptive predictors are needed to study historical behavior and forecast future state in various scenarios. In this paper, inspired by human attention mechanism and decomposition and reconstruction framework, we proposed a convolution-based dual-stage attention (CDA) architecture combined with Long Short-Term Memory networks (LSTM) for univariate time series forecasting. Specifically, we first use the decomposition algorithm to generate derived variables from target series. Input variables are then fed into the CDA-LSTM machine for further forecasting. In the Encoder–Decoder phase, for the first stage, attention operation is combined with the LSTM acting as an encoder, which could adaptively learn the relevant derived series to the target. In the second stage, the temporal attention mechanism is integrated with decoder aiming to automatically select the relevant encoder hidden states across all time steps. A convolution phase is concatenated parallelly to the Encoder–Decoder phase to reuse the historical information of the target and extract the mutation features. The experimental results demonstrate the proposed method could be adopted as expert systems for forecasting in multiple scenarios, and the superiority is verified by comparing with twelve baseline models on ten datasets. The practicability of different decomposition algorithms and convolution architectures is also discussed by extensive experiment. Overall, our work carries a significant value not merely in adaptive modeling of deep learning in time series issues, but also in the field of univariate data processing and prediction. [ABSTRACT FROM AUTHOR]

Cloud computing helps users to store and retrieve their data in the cloud online on an as-per-pay basis anytime, anywhere in the world. As a consequence, the security of data stored in the cloud serves as a key concern for cloud consumers due to ongoing hacking incidents in the cloud. Both cloud service providers (CSPs) and consumers face various attacks that pose a serious threat to both the organization and the user. Therefore, a robust intrusion detection system (IDS) needs to be configured to identify and avoid potential attacks in the cloud at an earlier stage. The state-of-the-art IDS system which used shallow machine learning techniques suffered from poor accuracy with a higher false alarm rate and higher response time. To overcome this problem, a hybrid semantic deep learning (HSDL) architecture is developed in this paper by integrating the long short-term memory (LSTM), convolutional neural network (CNN), and support vector machine (SVM) architectures. The semantic information present in the network traffic is identified using a semantic layer known as the Word2Vec embedding layer. The HSDL model classifies the intrusion present in the text along with its corresponding attack class. The normal text without any trace of intrusion is encrypted using an advanced encryption standard (AES) algorithm to enhance cloud storage security, and the optimal key with the largest key breaking time for the AES algorithm is selected using the crossover-based mine blast optimization algorithm (CMBA). The proposed HSDL scheme is evaluated and tested using two real-time intrusion detection benchmark datasets, namely NSL-KDD and UNSW-NB15. The proposed model achieves an accuracy of 99.98% for the NSL-KDD and 98.47% for the UNSW-NB15 dataset, respectively. The experimental and security analysis conducted proves the efficiency and robustness of the proposed scheme. [ABSTRACT FROM AUTHOR]

Traumatic Brain Injury (TBI) leads to intracranial hemorrhages (ICH), which is a severe illness resulted in death if it is not properly diagnosed and treated in the earlier stage. Presently, computer tomography (CT) images are widely used by radiologists to identify and locate the regions of ICH. But it is a tedious task and mainly depends on the professional radiologists. This paper develops new deep learning (DL)-based ICH diagnosis and classification (DL-ICH) model using optimal image segmentation with Inception Network. The proposed DL-ICH model involves preprocessing, segmentation, feature extraction, and classification. Firstly, the input data undergo format conversion where the NIfTI files are converted into JPEG format. Next, Kapur's thresholding with an elephant herd optimization (EHO) algorithm called KT-EHO is employed for image segmentation. Then, DL based Inception v4 network is applied as a feature extractor to extract a useful set of features and a multilayer perceptron (MLP) is used for classification. An extensive set of simulations takes place to ensure the effective diagnostic performance of the DL-ICH model and the results are investigated under diverse dimensions. The experimental results achieved better accuracy rate. [ABSTRACT FROM AUTHOR]

Person re-identification (Re-ID) across multi-camera views remains a very challenging task. In fact, the use of global convolutional neural network (CNN) features is not optimal since the appearance similarity of identities is overlooked. Hence, persons with similar global appearance (wearing in very similar clothes) may be confused and hard to distinguish. To address these shared challenges, some attribute labels (such as hat or bag) can improve the discriminative ability of Re-ID model by depicting person images with more detailed descriptions. Hence, we propose in this paper a new attributes recognition model, called ARN, based on a deep CNN model. Then, we propose a new approach based on attributes similarity and k-reciprocal re-ranking algorithm in order to enhance person Re-ID baseline performances. The proposed approach, called K-RNNA, is motivated first of all, by the potential of re-ranking methods to resort the 'baseline' re-identification results from the initial retrieval. Second, attributes, as a mid-level semantic representation of the person's local appearance, could also potentially resolve these challenges. In fact, learned semantic attributes such as gender, hair, accessories (hat, bags, backpack, etc.) are used as a supplement during the re-ranking phase in order to discriminate between identities with similar global appearance. The experimental study conducted on Market-1501, Duke-MTMC-reID and CUHK03 datasets confirms that when attributes are used as supplement in the verification phase, we improve considerably the person Re-ID performances and we reach competitive results in terms of mean average precision (mAP) and Rank-1. [ABSTRACT FROM AUTHOR]

Agricultural machines (AMs) refer to equipment usually used in agriculture such as tractors, hand tools, and power tools. It reduces the labor work, increases farms produce, enhances goods quality, and reduces farming time and cost-saving. However, the faults in the fuel system, blades, engine of the AM will often result in degraded vehicle performance, compromising the vehicle's efficiency and strength. To overcome these problems, fault detection algorithms are developed to identify the faults even before they occur with high classification accuracy. The deep convolutional neural network (DCNN) is a popular deep learning model that offers a high classification recognition rate, and it is widely adopted in similar fields for monitoring the health status of machines. Very few state-of-the-art works are available to identify the health state of agricultural machines using deep learning techniques and extracting the acoustic features from an audio recording. The acoustic signal-based agricultural machine health monitoring and fault prediction model using smartphones is a cost-effective option that is deployed in this proposed work. To optimize the network structure of the DCNN, this paper proposes a Levy flight optimization algorithm (LFOA). The DCNN-LFOA model is implemented on the smartphone's on-board device (OBD) along with the health monitoring application. The LFOA algorithm minimizes the number of neurons in the DCNN hidden layer and the number of input features from the audio recordings and enhances the classification accuracy. The LFOA algorithm provides the optimal solution which is essential in developing a lightweight DCNN model to implement in the edge processor (smartphone). The experimental results prove that the proposed model gives improved accuracy for the six faults to be classified and serves as a new research model to identify the health condition of the vehicles. [ABSTRACT FROM AUTHOR]

Computer-aided diagnosis system is becoming a more and more important tool in clinical treatment, which can provide a verification of the doctors' decisions. In this paper, we proposed a novel abnormal brain detection method for magnetic resonance image. Firstly, a pre-trained AlexNet was modified with batch normalization layers and trained on our brain images. Then, the last several layers were replaced with an extreme learning machine. A searching method was proposed to find the best number of layers to be replaced. Finally, the extreme learning machine was optimized by chaotic bat algorithm to obtain better classification performance. Experiment results based on 5 × hold-out validation revealed that our method achieved state-of-the-art performance. [ABSTRACT FROM AUTHOR]

With recently successful applications of deep learning in computer vision and general signal processing, deep learning has shown many unique advantages in medical signal processing. However, data labelling quality has become one of the most significant issues for AI applications, especially when it requires domain knowledge (e.g. medical image labelling). Besides, there might be noisy labels in practical datasets, which impairs the training process of neural networks. In this paper, we propose a semi-supervised algorithm for training data samples with noisy labels by performing selected positive learning and negative learning. To verify the effectiveness of the proposed scheme, we designed a portable ECG patch-RealCare and applied the algorithm on a real-life dataset. Our experimental results show that we can achieve an accuracy of 91.0%, which is 6.2% higher than a normal training process with ResNet. There are 65 patients in our dataset, and we randomly picked 2 patients to perform validation. [ABSTRACT FROM AUTHOR]

Vehicle positioning and vehicle identification of natural scene images are an important part of intelligent transportation systems and unmanned driving research. In current situation, there are still some problems in vehicle intelligent wireless positioning. In order to improve the intelligent wireless positioning efficiency of vehicles, based on the convolutional neural network, this research combines the concept of deep learning to carry out algorithm innovation in the research. Moreover, this paper combines the actual vehicle positioning problem points to collect data, simulates the vehicle positioning situation in a variety of complex situations, and designs a controlled test to verify. The results show that the algorithm of this study has certain effects, which can provide reference for subsequent related research and has certain practical significance. [ABSTRACT FROM AUTHOR]

Device-free passive wireless indoor localization is attracting great interest in recent years due to the widespread deployment of Wi-Fi devices and the numerous location-based services requirements. In this paper, we propose DFPhaseFL, the first device-free fingerprinting indoor localization system that purely uses CSI phase information. It utilizes the CSI phase information extracted from simply a single link to estimate the location of the target, neither requiring the target to wear any electronic equipment nor deploying a large number of access points and monitor devices. In DFPhaseFL, the raw CSI phases are extracted from the CSI measurements through the three antennas of the Intel WiFi Link 5300 wireless Network Interface Card (IWL 5300 NIC) firstly. Then, linear transformation and noise filtering are applied to acquire the calibrated CSI phases. Through experimental observations, we find that the calibrated CSI phase owns an unpredictable characteristic over time. Thus, it cannot be directly applied as a fingerprint. To this end, a transfer deep supervised neural network method combining deep neural network and transfer learning is proposed to obtain feature representations with both transferability and discriminability as fingerprints. Then, the DFPhaseFL system uses the SVM algorithm to obtain the estimation of the target location online. Experiment results demonstrate that the DFPhaseFL owns a better estimation precision compared with the other state of art, and maintain a stable localization accuracy for a long time without reacquiring the fingerprint database. [ABSTRACT FROM AUTHOR]

In hyperspectral image (HSI) classification, a big challenge is the limited sample size with a relatively high feature dimension. Therefore, effective feature extraction of data is essential, which is desired to remove the redundancy as well as improve the discrimination. A huge number of methods have been proposed for HSI feature extraction. In recent years, deep learning-based feature extraction algorithms have shown their superiorities in various classification problems. Within them, deep PCA (DPCA) is a simple but efficient algorithm, which runs fast due to the absence of back-propagation. However, DPCA fails to provide satisfactory classification accuracies on HSI datasets. In this paper, we try to combine DPCA with residual-based multi-scale feature extraction and propose a residual deep PCA (RDPCA) feature extraction algorithm for HSI classification. It is a hierarchical approach consisting of multiple layers. Within each layer, PCA is utilized for layer-wise feature extraction, and the reconstruction residual is fed into the next layer. When the feature is passed deeper into the RDPCA network, finer details are mined. The layer-wise features are concatenated to form the final output feature. Furthermore, to enhance the ability of nonlinear feature extraction, we add activation functions between adjacent layers. Experimental results on real-world HSI datasets have shown the superiority of the proposed RDPCA over DPCA and PCA. [ABSTRACT FROM AUTHOR]

Adversarial examples are artificially crafted to mislead deep learning systems into making wrong decisions. In the research of attack algorithms against multi-class image classifiers, an improved strategy of applying category explanation to the generation control of targeted adversarial example is proposed to reduce the perturbation noise and improve the adversarial robustness. On the basis of C&W adversarial attack algorithm, the method uses Grad-Cam, a category visualization explanation algorithm of CNN, to dynamically obtain the salient regions according to the signal features of source and target categories during the iterative generation process. The adversarial example of non-global perturbation is finally achieved by gradually shielding the non-salient regions and fine-tuning the perturbation signals. Compared with other similar algorithms under the same conditions, the method enhances the effects of the original image category signal on the perturbation position. And it makes up for the shortcomings of the adversarial example algorithm in terms of interpretability and teaching intuitiveness. Experimental results show that the improved adversarial examples have higher PSNR. In addition, in a variety of different defense processing tests, the examples can keep high adversarial performance and show strong attacking robustness. [ABSTRACT FROM AUTHOR]

Traffic sensing system requires to periodically collect spatial–temporal traffic data distributed among road networks, which results in overhigh bandwidth consumption and storage cost in a large-scale road network. Several compressive sensing-based algorithms are proposed to reconstruct missing traffic data with limited traffic observation. However, there still exist great challenges to be addressed. First, these existing algorithms are always iteration-based, whose time complexity will explosively increase with the growth of network scale. Furthermore, these algorithms have to be re-executed even if only a small fraction of data changes, which is not suitable for dynamic traffic environments. To overcome these issues, we investigate a novel service architecture of traffic sensing based on mobile edge computing where collected data is pre-processed at the edge node and reconstructed at cloud servers, respectively. On this basis, we formulate the problem of Missing Traffic Data reconstruction (MTDR), which aims at maximizing data reconstruction accuracy within limited observation data. Further, we develop a deep-learning-based algorithm called stacked denoising autoencoder for MTDR (SDAE-MTDR), where three denoising autoencoders are trained in order and then stacked together for parameter fine-tuning based on cross-entropy-based loss function. Finally, we conduct comprehensive performance evaluation based on realistic vehicular traces and the simulation results demonstrate the superiority of the proposed algorithm compared with competitive solutions. [ABSTRACT FROM AUTHOR]

Health monitoring, rehabilitation, and fitness are just a few domains where human activity recognition can be applied. In this study, a deep learning approach has been proposed to recognise ambulation and fitness activities from data collected by five participants using smart insoles. Smart insoles, consisting of pressure and inertial sensors, allowed for seamless data collection while minimising user discomfort, laying the baseline for the development of a monitoring and/or rehabilitation system for everyday life. The key objective has been to enhance the deep learning model performance through several techniques, including data segmentation with overlapping technique (2 s with 50% overlap), signal down-sampling by averaging contiguous samples, and a cost-sensitive re-weighting strategy for the loss function for handling the imbalanced dataset. The proposed solution achieved an Accuracy and F1-Score of 98.56% and 98.57%, respectively. The Sitting activities obtained the highest degree of recognition, closely followed by the Spinning Bike class, but fitness activities were recognised at a higher rate than ambulation activities. A comparative analysis was carried out both to determine the impact that pre-processing had on the proposed core architecture and to compare the proposed solution with existing state-of-the-art solutions. The results, in addition to demonstrating how deep learning solutions outperformed those of shallow machine learning, showed that in our solution the use of data pre-processing increased performance by about 2%, optimising the handling of the imbalanced dataset and allowing a relatively simple network to outperform more complex networks, reducing the computational impact required for such applications. [ABSTRACT FROM AUTHOR]

Multiple-object tracking is a fundamental computer vision task which is gaining increasing attention due to its academic and commercial potential. Multiple-object detection, recognition and tracking are quite desired in many domains and applications. However, accurate object tracking is very challenging, and things are even more challenging when multiple objects are involved. The main challenges that multiple-object tracking is facing include the similarity and the high density of detected objects, while also occlusions and viewpoint changes can occur as the objects move. In this article, we introduce a real-time multiple-object tracking framework that is based on a modified version of the Deep SORT algorithm. The modification concerns the process of the initialization of the objects, and its rationale is to consider an object as tracked if it is detected in a set of previous frames. The modified Deep SORT is coupled with YOLO detection methods, and a concrete and multi-dimensional analysis of the performance of the framework is performed in the context of real-time multiple tracking of vehicles and pedestrians in various traffic videos from datasets and various real-world footage. The results are quite interesting and highlight that our framework has very good performance and that the improvements on Deep SORT algorithm are functional. Lastly, we show improved detection and execution performance by custom training YOLO on the UA-DETRAC dataset and provide a new vehicle dataset consisting of 7 scenes, 11.025 frames and 25.193 bounding boxes. [ABSTRACT FROM AUTHOR]

With the rapid growth of high-voltage transmission lines, the number of power transmission line equipments is correspondingly increasing. Power insulator is the basic component which plays the key role in the stable operation of power system. As a common defect of power insulators, missing-cap issue will affect the structural strength and durability of different power insulators. Therefore, the condition monitoring of power insulators is a daily but priority power line inspection task. Faced with the weak image features of small insulator defects in the aerial images, the conventional handcrafted features could not extract effectively powerful image features. Meanwhile, the small-scale insulator defects will bring a certain effect to the model training of deep learning. Therefore, the high-efficiency and accurate defect inspection still present a challenging task against complex backgrounds. To address the above issues, aimed at the missing-cap defects of power insulators, a novel defect identification algorithm from aerial images is proposed by taking advantage of state-of-the-art deep learning and transfer learning models. Fused with Spatial Pyramid Pooling (SPP) and MobileNet networks, a light deep convolutional neural network (DCNN) model based on You Only Look Once (YOLO) V3 network is proposed for fast and accurate insulator location to remove complex background interference. On the basis, combined with Dempster–Shafer (DS) evidence theory, the improved transfer learning model based on feature fusion is proposed for high-precision defect identification of power insulators. Experiments show that the proposed method could acquire a better identification performance against complex power inspection environment compared with other related detection models. [ABSTRACT FROM AUTHOR]

Experience replay (ER) improves the data efficiency of off-policy reinforcement learning (RL) algorithms by allowing an agent to store and reuse its past experiences in a replay buffer. While many techniques have been proposed to enhance ER by biasing how experiences are sampled from the buffer, thus far they have not considered strategies for refreshing experiences inside the buffer. In this work, we introduce LucidDreaming forExperienceReplay (LiDER), a conceptually new framework that allows replay experiences to be refreshed by leveraging the agent's current policy. LiDER consists of three steps: First, LiDER moves an agent back to a past state. Second, from that state, LiDER then lets the agent execute a sequence of actions by following its current policy—as if the agent were "dreaming" about the past and can try out different behaviors to encounter new experiences in the dream. Third, LiDER stores and reuses the new experience if it turned out better than what the agent previously experienced, i.e., to refresh its memories. LiDER is designed to be easily incorporated into off-policy, multi-worker RL algorithms that use ER; we present in this work a case study of applying LiDER to an actor–critic-based algorithm. Results show LiDER consistently improves performance over the baseline in six Atari 2600 games. Our open-source implementation of LiDER and the data used to generate all plots in this work are available at https://github.com/duyunshu/lucid-dreaming-for-exp-replay. [ABSTRACT FROM AUTHOR]

Clustering of face images is a well-known clustering problem with several applications as for example the automated face tagging in photo albums. The performance of face clustering algorithms largely depends on the face representations that are used. Therefore, the exploitation of the internal face representation (embedding) obtained by feeding the face image to a deep face network has led to increased clustering performance. In this work, we perform face clustering by combining multiple representations of each image obtained from several deep face networks. The multiple deep representations are not exploited in a straightforward manner (simple vector concatenation). Instead a weight is assigned to each representation to reflect its importance on the clustering result and a weighted clustering algorithm is employed to automatically adjust the weights and, consequently, the contribution of each representation on the clustering solution. It should also be noted that in the proposed approach, the number of clusters (number of faces) is not given as input, but it is automatically estimated using the silhouette criterion. The conducted experiments have shown that the weighted combination of deep face representations leads to improved clustering performance compared to using single representations. [ABSTRACT FROM AUTHOR]

A recent research area in unsupervised learning is the combination of representation learning with deep neural networks and data clustering. The success of deep learning for supervised tasks is widely established. However, recent research has demonstrated how neural networks are able to learn representations to improve clustering in their intermediate feature space, using specific regularizations. By considering representation learning and clustering as a joint task, models learn clustering-friendly spaces and outperform two-stage approaches where dimensionality reduction and clustering are performed separately. Recently, this idea has been extended to topology-preserving clustering models, known as self-organizing maps (SOM). This work is a thorough study on the deep embedded self-organizing map (DESOM), a model composed of an autoencoder and a SOM layer, training jointly the code vectors and network weights to learn SOM-friendly representations. In other words, SOM induces a form a regularization to improve the quality of quantization and topology in latent space. After detailing the architecture, loss and training algorithm, we study hyperparameters with a series of experiments. Different SOM-based models are evaluated in terms of clustering, visualization and classification on benchmark datasets. We study benefits and trade-offs of joint representation learning and self-organization. DESOM achieves competitive results, requires no pretraining and produces topologically organized visualizations. [ABSTRACT FROM AUTHOR]

An increasing number of complex problems have naturally posed significant challenges in decision-making theory and reinforcement learning practices. These problems often involve multiple conflicting reward signals that inherently cause agents' poor exploration in seeking a specific goal. In extreme cases, the agent gets stuck in a sub-optimal solution and starts behaving harmfully. To overcome such obstacles, we introduce two actor-critic deep reinforcement learning methods, namely Multi-Critic Single Policy (MCSP) and Single Critic Multi-Policy (SCMP), which can adjust agent behaviors to efficiently achieve a designated goal by adopting a weighted-sum scalarization of different objective functions. In particular, MCSP creates a human-centric policy that corresponds to a predefined priority weight of different objectives. Whereas, SCMP is capable of generating a mixed policy based on a set of priority weights, i.e., the generated policy uses the knowledge of different policies (each policy corresponds to a priority weight) to dynamically prioritize objectives in real time. We examine our methods by using the Asynchronous Advantage Actor-Critic (A3C) algorithm to utilize the multithreading mechanism for dynamically balancing training intensity of different policies into a single network. Finally, simulation results show that MCSP and SCMP significantly outperform A3C with respect to the mean of total rewards in two complex problems: Food Collector and Seaquest. [ABSTRACT FROM AUTHOR]

Deep learning-based recommender systems have gained much attention due to the advantage of encoding content-based information, such as user textual reviews and item descriptions, images, or videos, without the trouble of manually crafting feature vectors. However, those systems are trained from scratch with randomly initialized parameters, where the training process can take a long time to converge. With the most recent breakthroughs in Natural Language Processing using transfer learning, pre-trained transformer-based models now provide a better foundation for textual information encoding. This inspires us to propose a transformer-based recommender system using transfer learning. As the first core contribution in this work, we apply transfer learning to the system, by fine-tuning the pre-trained transformer models for information encoding. The experiment result shows that the proposed system outperforms several other deep learning-based recommender systems on multiple datasets. As the second core contribution, we propose a novel user vector encoding algorithm that assists all the models to achieve a better performance, when the user content information is not available. [ABSTRACT FROM AUTHOR]

For multi-label learning, the specific features are extracted from the instances under the supervised of class label is meaningful, and the "purified" feature representation can also be shared with other features during learning process. Besides, it is essential to distinguish the inter-instance relations in input space and inter-label correlation relations in the output space on the multi-label datasets, which is conducive to improve the performance of the multi-label algorithm. However, most current multi-label algorithms aim to capture the mapping between instances and labels, while ignoring the information about instance relations and label correlations in the multi-label data structure. Motivated by these issues, we leverage the deep network to learn the special feature representations for multi-label components without abandoning overlapped features which may belong to other multi-label components. Meanwhile, the Euclidean matrices are leveraged to construct the diagonal matrix for the diffusion function, obtaining the new class latent representation by graph-based diffusion method preserve the inter-instance relations; it ensures that similar features have similar label sets. Further, considering that the contributions of these feature representation are different and have distinct influences on the final multi-label prediction results, the self-attention mechanism is introduced to fusion the other label-specific instance features to build the new joint feature representation, which derives dynamic weights for multi-label prediction. Finally, experimental results on the real data sets show promising wide availability for our approach. [ABSTRACT FROM AUTHOR]

During the last decade, a significant research progress has been drawn in both the theoretical aspects and the applications of Deep Learning Neural Networks. Besides their spectacular applications, optimal architectures of these neural networks may speed up the learning process and exhibit better generalization results. So far, many growing and pruning algorithms have been proposed by many researchers to deal with the optimization of standard Feedforward Neural Network architectures. However, applying both the growing and the pruning on the same net may lead a good model for a big data set and hence good selection results. This work is devoted to propose a new Growing and pruning Learning algorithm for Deep Neural Networks. This new algorithm is presented and applied on diverse medical data sets. It is shown that this algorithm outperforms various other artificial intelligent techniques in terms of accuracy and simplicity of the resulting architecture. [ABSTRACT FROM AUTHOR]

The article title " I A novel equilibrium optimization algorithm for multi-thresholding image segmentation problems i " by Mohamed Abdel-Basset, Victor Chang and Reda Mohamed presented an equilibrium optimization algorithm to find the optimal threshold value for a grayscale image. Hybrid intelligent methods have proven their efficiency to solve NP-hard optimization problems. The performance of the proposed algorithm was compared against the other popular optimization algorithms such as whale optimization algorithm, bat algorithm, sine-cosine algorithm, salp swarm algorithm, Harris hawk's algorithm, crow search algorithm, and particle swarm optimization. [Extracted from the article]

In this study we represent malware as opcode sequences and detect it using a deep belief network (DBN). Compared with traditional shallow neural networks, DBNs can use unlabeled data to pretrain a multi-layer generative model, which can better represent the characteristics of data samples. We compare the performance of DBNs with that of three baseline malware detection models, which use support vector machines, decision trees, and the k-nearest neighbor algorithm as classifiers. The experiments demonstrate that the DBN model provides more accurate detection than the baseline models. When additional unlabeled data are used for DBN pretraining, the DBNs perform better than the other detection models. We also use the DBNs as an autoencoder to extract the feature vectors of executables. The experiments indicate that the autoencoder can effectively model the underlying structure of input data and significantly reduce the dimensions of feature vectors. [ABSTRACT FROM AUTHOR]