Search

The task of fine-grained named entity recognition is to locate entities in text and classify them into predefined fine-grained categories. At present, Chinese fine-grained NER only uses the pretrained language model to encode the characters in the sentence and lacks the ability to extract the deep semantic, sequence, and position information. The sequence annotation method is character-based and lacks the processing of entity boundaries. Fine-grained entity categories have a high degree of similarity, which makes it difficult to distinguish similar categories. To solve the above problems, this paper constructs the BILTAR deep semantic extraction module and adds the GlobalPointer module to improve the accuracy of Chinese fine-grained named entity recognition. The BILTAR module is used to extract deep semantic features from the coding information of pretrained language models and use higher-quality features to improve the model performance. In the GlobalPointer module, the model first adds the rotation position encoding information to the feature vector, using the position information to achieve data enhancement. Finally, the model considers all possible entity boundaries through the GlobalPointer module and calculates the scores for all possible entity boundaries in each category. In this paper, all possible entity boundaries in the text are considered by the above method, and the accuracy of entity recognition is improved. In this paper, the corresponding experiments were carried out on CLUENER 2020 and the micro Chinese fine-grained NER dataset, and the F1 scores of the model in this paper reached 80.848% and 75.751%, respectively. In ablation experiments, the proposed method outperforms the most advanced baseline model and improves the performance of the basic model.

Deep neural networks (DNNs) have successfully delivered cutting-edge performance in several fields. With the broader deployment of DNN models on critical applications, the security of DNNs has become an active and yet nascent area. Attacks against DNNs can have catastrophic results, according to recent studies. Poisoning attacks, including backdoor attacks and Trojan attacks, are one of the growing threats against DNNs. Having a wide-angle view of these evolving threats is essential to better understand the security issues. In this regard, creating a semantic model and a knowledge graph for poisoning attacks can reveal the relationships between attacks across intricate data to enhance the security knowledge landscape. In this paper, we propose a DNN poisoning attack ontology (DNNPAO) that would enhance knowledge sharing and enable further advancements in the field. To do so, we have performed a systematic review of the relevant literature to identify the current state. We collected 28,469 papers from the IEEE, ScienceDirect, Web of Science, and Scopus databases, and from these papers, 712 research papers were screened in a rigorous process, and 55 poisoning attacks in DNNs were identified and classified. We extracted a taxonomy of the poisoning attacks as a scheme to develop DNNPAO. Subsequently, we used DNNPAO as a framework by which to create a knowledge base. Our findings open new lines of research within the field of AI security.

The remarkable results of applying machine learning algorithms to complex tasks are well known. They open wide opportunities in natural language processing, image recognition, and predictive analysis. However, their use in low-power intelligent systems is restricted because of high computational complexity and memory requirements. This group includes a wide variety of devices, from smartphones and Internet of Things (IoT)smart sensors to unmanned aerial vehicles (UAVs), self-driving cars, and nodes of Edge Computing systems. All of these devices have severe limitations to their weight and power consumption. To apply neural networks in these systems efficiently, specialized hardware accelerators are used. However, hardware implementation of some neural network operations is a challenging task. Sigmoid activation is popular in the classification problem and is a notable example of such a complex operation because it uses division and exponentiation. The paper proposes efficient implementations of this activation for dynamically reconfigurable accelerators. Reconfigurable computing environments (RCE) allow achieving reconfigurability of accelerators. The paper shows the advantages of applying such accelerators in low-power systems, proposes the centralized and distributed hardware implementations of the sigmoid, presents comparisons with the results of other studies, and describes application of the proposed approaches to other activation functions. Timing simulations of the developed Verilog modules show low delay (14–18.5 ns) with acceptable accuracy (average absolute error is 4 × 10−3).

This paper describes a post-evaluation analysis of the system developed by ViVoLAB research group for the IberSPEECH-RTVE 2020 Multimodal Diarization (MD) Challenge. This challenge focuses on the study of multimodal systems for the diarization of audiovisual files and the assignment of an identity to each segment where a person is detected. In this work, we implemented two different subsystems to address this task using the audio and the video from audiovisual files separately. To develop our subsystems, we used the state-of-the-art speaker and face verification embeddings extracted from publicly available deep neural networks (DNN). Different clustering techniques were also employed in combination with the tracking and identity assignment process. Furthermore, we included a novel back-end approach in the face verification subsystem to train an enrollment model for each identity, which we have previously shown to improve the results compared to the average of the enrollment data. Using this approach, we trained a learnable vector to represent each enrollment character. The loss function employed to train this vector was an approximated version of the detection cost function (aDCF) which is inspired by the DCF widely used metric to measure performance in verification tasks. In this paper, we also focused on exploring and analyzing the effect of training this vector with several configurations of this objective loss function. This analysis allows us to assess the impact of the configuration parameters of the loss in the amount and type of errors produced by the system.

One of the central aspects of science is systematic problem-solving. Therefore, problem and solution statements are an integral component of the scientific discourse. The scientific analysis would be more successful if the problem–solution claims in scientific texts were automatically classified. It would help in knowledge mining, idea generation, and information classification from scientific texts. It would also help to compare scientific papers and automatically generate review articles in a given field. However, computational research on problem–solution patterns has been scarce. The linguistic analysis, instructional-design research, theory, and empirical methods have not paid enough attention to the study of problem–solution patterns. This paper tries to solve this issue by applying the computational techniques of machine learning classifiers and neural networks to a set of features to intelligently classify a problem phrase from a non-problem phrase and a solution phrase from a non-solution phrase. Our analysis shows that deep learning networks outperform machine learning classifiers. Our best model was able to classify a problem phrase from a non-problem phrase with an accuracy of 90.0% and a solution phrase from a non-solution phrase with an accuracy of 86.0%.

Mainstream transferable adversarial attacks tend to introduce noticeable artifacts into the generated adversarial examples, which will impair the invisibility of adversarial perturbation and make these attacks less practical in real-world scenarios. To deal with this problem, in this paper, we propose a novel black-box adversarial attack method that can significantly improve the invisibility of adversarial examples. We analyze the sensitivity of a deep neural network in the frequency domain and take into account the characteristics of the human visual system in order to quantify the contribution of each frequency component in adversarial perturbation. Then, we collect a set of candidate frequency components that are insensitive to the human visual system by applying K-means clustering and we propose a joint loss function during the generation of adversarial examples, limiting the frequency distribution of perturbations during attacks. The experimental results show that the proposed method significantly outperforms existing transferable black-box adversarial attack methods in terms of invisibility, which verifies the superiority, applicability and potential of this work.

With the extensive adoption of generative models across various domains, the protection of copyright for these models has become increasingly vital. Some researchers suggest embedding watermarks in the images generated by these models as a means of preserving IP rights. In this paper, we find that existing generative model watermarking introduces high-frequency artifacts in the high-frequency spectrum of the marked images, thereby compromising the imperceptibility and security of the generative model watermarking system. Given this revelation, we propose an innovative image watermarking technology that takes into account frequency-domain imperceptibility. Our approach abandons the conventional convolutional neural network (CNN) structure typically used as the watermarking embedding network in popular watermarking techniques. This helps the image watermarking system to avoid the inherent high-frequency artifacts commonly produced by CNNs. In addition, we design a frequency perturbation generation network to generate low-frequency perturbations. These perturbations are subsequently added as watermarks to the low-frequency components of the carrier image, thus minimizing the impact of the watermark embedding process on the high-frequency properties of the image. The results show that our proposed watermarking framework can effectively embed low-frequency perturbation watermarks into images and effectively suppress high-frequency artifacts in images, thus significantly improving the frequency-domain imperceptibility and security of the image watermarking system. The introduced approach enhances the average invisibility performance in the frequency domain by up to 24.9% when contrasted with prior methods. Moreover, the method attains superior image quality (>50 dB) in the spatial domain, accompanied by a 100% success rate in watermark extraction in the absence of attacks. This underscores its capability to uphold the efficacy of the protected network and preserve the integrity of the watermarking process. It always maintains excellent imperceptibility and robustness. Thus, the framework shows great potential as a state-of-the-art solution for protecting intellectual property.

Three-dimensional (3D) models provide the most intuitive representation of geological conditions. Traditional modeling methods heavily depend on technicians’ expertise and lack ease of updating. In this study, we introduce a deep learning-based method for 3D geological implicit modeling, leveraging a substantial dataset of geological drilling data. By applying resampling and normalization techniques, we standardize drilling data and significantly expand the dataset, making it suitable for training deep neural networks. Utilizing the characteristics of the sample data, we design and establish the network structure, loss function, and parameter configurations, resulting in the training of a deep neural network with high accuracy and robust generalization capability. Ultimately, we utilize the dataset generated from the network’s predictions to render and construct the 3D geological model. The research in this paper demonstrates the significant promise of deep neural networks in addressing geological challenges. The deep learning-based implicit 3D modeling method surpasses traditional approaches in terms of generalization, convenience, and adaptability.

Open-domain question answering requires the task of retrieving documents with high relevance to the query from a large-scale corpus. Deep learning-based dense retrieval methods have become the primary approach for finding related documents. Although deep learning-based methods have improved search accuracy compared to traditional techniques, they simultaneously impose a considerable increase in computational burden. Consequently, research on efficient models and methods that optimize the trade-off between search accuracy and time to alleviate computational demands is required. In this paper, we propose a Korean document retrieval method utilizing ColBERT’s late interaction paradigm to efficiently calculate the relevance between questions and documents. For open-domain Korean question answering document retrieval, we construct a Korean dataset using various corpora from AI-Hub. We conduct experiments comparing the search accuracy and inference time among the traditional IR (information retrieval) model BM25, the dense retrieval approach utilizing BERT-based models for Korean, and our proposed method. The experimental results demonstrate that our approach achieves a higher accuracy than BM25 and requires less search time than the dense retrieval method employing KoBERT. Moreover, the most outstanding performance is observed when using KoSBERT, a pre-trained Korean language model that learned to position semantically similar sentences closely in vector space.

Congestive heart failure carries immense importance in the realm of public health. This significance arises from its substantial influence on the number of lives lost, economic burdens, the potential for prevention, and the opportunity to enhance the well-being of both individuals and the broader community through decision-making in healthcare. Several researchers have proposed neural networks for classification of different congestive heart failure categories. However, there is little information about the confidence of the prediction on short-term series. Therefore, evaluating classification models is required for effective decision-making in healthcare. This paper explores the use of three classical variants of neural networks to classify three groups of patients with congestive heart failure. The study considered the iterative method Multilayer Perceptron neural network (MLP), two non-iterative models (Extreme Learning Machine (ELM) and Random Vector Functional Link Network (RVFL)), and the CNN approach. The results showed that the deep feature learning system obtained better classification rates than MLP, ELM, and RVFL. Several scenarios designed by coupling some deep feature maps with the RVFL and MLP models showed very high simulation accuracy. The overall accuracy rate of CNN–MLP and CNN–RVFL varies between 98% and 99%.

Remote sensing involves actions to obtain information about an area located on Earth. In the Amazon region, the presence of clouds is a common occurrence, and the visualization of important terrestrial information in the image, like vegetation and temperature, can be difficult. In order to estimate land surface temperature (LST) and the normalized difference vegetation index (NDVI) from satellite images with cloud coverage, the inpainting approach will be applied to remove clouds and restore the image of the removed region. This paper proposes the use of the neural network LaMa (large mask inpainting) and the scalable model named Big LaMa for the automatic reconstruction process in satellite images. Experiments are conducted on Landsat-8 satellite images of the Amazon rainforest in the state of Acre, Brazil. To evaluate the architecture’s accuracy, the RMSE (root mean squared error), SSIM (structural similarity index) and PSNR (peak signal-to-noise ratio) metrics were used. The LST and NDVI of the reconstructed image were calculated and compared qualitatively and quantitatively, using scatter plots and the chosen metrics, respectively. The experimental results show that the Big LaMa architecture performs more effectively and robustly in restoring images in terms of visual quality. And the LaMa network shows minimal superiority for the measured metrics when addressing medium marked areas. When comparing the results achieved in NDVI and LST of the reconstructed images with real cloud coverage, great visual results were obtained with Big LaMa.

It is crucial to analyze opinions about the significant shift in education systems around the world, because of the widespread use of e-learning, to gain insight into the state of education today. A particular focus should be placed on the feedback from students regarding the profound changes they experience when using e-learning. In this paper, we propose a model that combines fuzzy logic with bidirectional long short-term memory (BiLSTM) for the sentiment analysis of students’ textual feedback on e-learning. We obtained this feedback from students’ tweets expressing their opinions about e-learning. There were some ambiguous characteristics in terms of the writing style and language used in the collected feedback. It was written informally and not in adherence to standardized Arabic language writing rules by using the Saudi dialects. The proposed model benefits from the capabilities of the deep neural network BiLSTM to learn and also from the ability of fuzzy logic to handle uncertainties. The proposed models were evaluated using the appropriate evaluation metrics: accuracy, F1-score, precision, and recall. The results showed the effectiveness of our proposed model and that it worked well for analyzing opinions obtained from Arabic texts written in Saudi dialects. The proposed model outperformed the compared models by obtaining an accuracy of 86% and an F1-score of 85%.

In this paper, we propose an advanced method for adversarial training that focuses on leveraging the underlying structure of adversarial perturbation distributions. Unlike conventional adversarial training techniques that consider adversarial examples in isolation, our approach employs clustering algorithms in conjunction with dimensionality reduction techniques to group adversarial perturbations, effectively constructing a more intricate and structured feature space for model training. Our method incorporates density and boundary-aware clustering mechanisms to capture the inherent spatial relationships among adversarial examples. Furthermore, we introduce a strategy for utilizing adversarial perturbations to enhance the delineation between clusters, leading to the formation of more robust and compact clusters. To substantiate the method’s efficacy, we performed a comprehensive evaluation using well-established benchmarks, including MNIST and CIFAR-10 datasets. The performance metrics employed for the evaluation encompass the adversarial clean accuracy trade-off, demonstrating a significant improvement in both robust and standard test accuracy over traditional adversarial training methods. Through empirical experiments, we show that the proposed clustering-based adversarial training framework not only enhances the model’s robustness against a range of adversarial attacks, such as FGSM and PGD, but also improves generalization in clean data domains.

The use of continuous monitoring systems to control aspects such as noise pollution has grown in recent years. The commercial monitoring systems used to date only provide information on noise levels but do not identify the noise sources that generate them. The identification of noise sources is an important aspect in order to apply corrective measures to mitigate the noise levels. In this sense, new technological advances like machine listening can enable the addition of other capabilities to sound monitoring systems such as the detection and classification of noise sources. Despite the increasing development of these systems, researchers have to face some shortcomings. The most frequent ones are on the one hand, the lack of data recorded in real environments and on the other hand, the need for automatic labelling of large volumes of data collected by working monitoring systems. In order to address these needs, in this paper, we present our own sound database recorded in an urban environment. Some baseline results for the database are provided using two original convolutional neural network based sound events classification systems. Additionally, a state of the art transformer-based audio classification system (AST) has been applied to obtain some baseline results. Furthermore, the database has been used for evaluating a semi-supervised strategy to train a classifier for automatic labelling that can be refined by human labellers afterwards.

The paper introduces a range of efficient algorithmic solutions for implementing the fundamental filtering operation in convolutional layers of convolutional neural networks on fully parallel hardware. Specifically, these operations involve computing M inner products between neighbouring vectors generated by a sliding time window from the input data stream and an M-tap finite impulse response filter. By leveraging the factorisation of the Hankel matrix, we have successfully reduced the multiplicative complexity of the matrix-vector product calculation. This approach has been applied to develop fully parallel and resource-efficient algorithms for M values of 3, 5, 7, and 9. The fully parallel hardware implementation of our proposed algorithms achieves approximately a 30% reduction in embedded multipliers compared to the naive calculation methods.

Accurate classification and identification of Internet traffic are crucial for maintaining network security. However, unknown network traffic in the real world can affect the accuracy of current machine learning models, reducing the efficiency of traffic classification. Existing unknown traffic classification algorithms are unable to optimize traffic features and require the entire system to be retrained each time new traffic data are collected. This results in low recognition efficiency, making the algoritms unsuitable for real-time application detection. To solve the above issues, we suggest a multi-feature fusion-based incremental technique for detecting unknown traffic in this paper. The approach employs a multiple-channel parallel architecture to extract temporal and spatial traffic features. It then uses the mRMR algorithm to rank and fuse the features extracted from each channel to overcome the issue of redundant encrypted traffic features. In addition, we combine the density-ratio-based clustering algorithm to identify the unknown traffic features and update the model via incremental learning. The cassifier enables real-time classification of known and unknown traffic by learning newly acquired class knowledge. Our model can identify encrypted unknown Internet traffic with at least 86% accuracy in various scenarios, using the public ISCX-VPN-Tor datasets. Furthermore, it achieves 90% accuracy on the intrusion detection dataset NSL-KDD. In our self-collected dataset from a real-world environment, the accuracy of our model exceeds 96%. This work offers a novel method for identifying unknown network traffic, contributing to the security preservation of network environments.

For fresh table grapes, flavor is one of the most important components of their overall quality. The flavor of table grapes includes both their taste and aroma, involving multiple physical and chemical properties, such as soluble solids. In this paper, we investigate six factors, divide flavor ratings into a range of five grades based on the results of trained food tasters, and propose a deep-neural-network-based flavor evaluation model that integrates an attention mechanism. After training, the proposed model achieved a prediction accuracy of 94.8% with an average difference of 2.657 points between the predicted score and the actual score. This work provides a promising solution to the evaluation of table grapes and has the potential to improve product quality for future breeding in agricultural engineering.

A bite from a bug may expose the affected person to serious, life-threatening conditions, which may require immediate medical attention. The identification of the bug bite may be challenging even for experienced medical personnel due to the different manifestations of the bites and similarity to other skin conditions. This motivated our work on a computer-aided system that offers information on the bug bite based on the classification of bug bite images. Recently, there have been significant advances of methods for image classification for the detection of various skin conditions. However, there are very few sources that discuss the classification of bug bites. The goal of our research is to fill in this gap in the literature and offer a comprehensive approach for the analysis of this topic. This includes (1) the creation of a dataset that is larger than those considered in the related sources; (2) the exploration and analysis of the application of pre-trained state-of-the-art deep learning architectures with transfer learning, used in this study to overcome the challenges of low-size datasets and computational burden; (3) the further improvement of the classification performance of the individual CNNs by proposing an ensemble of models, and finally, (4) the implementation and description of an end-to-end system for bug bite classification from images taken with mobile phones, which should be beneficial to the medical personnel in the diagnostic process. In this paper, we give a detailed discussion of the models’ architecture, back-end architecture, and performance. According to the general evaluation metrics, DenseNet169 with an accuracy of 78% outperformed the other individual CNN models. However, the overall best performance (accuracy of 86%) was achieved by the proposed stacking ensemble model. These results are better than the results in the limited related work. Additionally, they show that deep CNNs and transfer learning can be successfully applied to the problem of the classification of bug bites.

As a self-supervised learning paradigm, contrastive learning has been widely used to pre-train a powerful encoder as an effective feature extractor for various downstream tasks. This process requires numerous unlabeled training data and computational resources, which makes the pre-trained encoder become the valuable intellectual property of the owner. However, the lack of a priori knowledge of downstream tasks makes it non-trivial to protect the intellectual property of the pre-trained encoder by applying conventional watermarking methods. To deal with this problem, in this paper, we introduce AWEncoder, an adversarial method for watermarking the pre-trained encoder in contrastive learning. First, as an adversarial perturbation, the watermark is generated by enforcing the training samples to be marked to deviate respective location and surround a randomly selected key image in the embedding space. Then, the watermark is embedded into the pre-trained encoder by further optimizing a joint loss function. As a result, the watermarked encoder not only performs very well for downstream tasks, but also enables us to verify its ownership by analyzing the discrepancy of output provided using the encoder as the backbone under both white-box and black-box conditions. Extensive experiments demonstrate that the proposed work enjoys quite good effectiveness and robustness on different contrastive learning algorithms and downstream tasks, which has verified the superiority and applicability of the proposed work.

Deep learning technology dominates current research in image denoising. However, denoising performance is limited by target noise feature loss from information propagation in association with the depth of the network. This paper proposes a Dense Residual Feature Extraction Network (DRFENet) combined with a Dense Enhancement Block (DEB), a Residual Dilated Block (RDB), a Feature Enhancement Block (FEB), and a Simultaneous Iterative Reconstruction Block (SIRB). The DEB uses our proposed interval transmission strategy to enhance the extraction of noise features in the initial stage of the network. The RDB module uses a combination strategy of concatenated dilated convolution and a skip connection, and the local features are amplified through different perceptual dimensions. The FEB enhances local feature information. The SIRB uses an attention block to learn the noise distribution while using residual learning (RL) technology to reconstruct a denoised image. The combination strategy in DRFENet makes the neural network deeper to obtain higher fine-grained image information. We respectively examined the performance of DRFENet in gray image denoising on datasets BSD68 and SET12 and color image denoising on datasets McMaster, Kodak24, and CBSD68. The experimental results showed that the denoising accuracy of DRFENet is better than most existing image-denoising methods under PSNR and SSIM evaluation indicators.

Display-to-camera (D2C) communications has emerged as a key method for next-generation videos that offer side information to camera-equipped devices during normal viewing. This paper presents Deep learning-driven Complementary Color Barcode-based Optical Camera Communications (DeepCCB-OCC), a D2C system using multiple deep neural networks built for imperceptible transmission and reliable communication in a D2C link. DeepCCB-OCC takes advantage of a the You Only Look Once (YOLO) model to provide seamless detection of a color barcode area in electronic displays. To identify transmitted color barcode symbols in the received image, we define various color barcode patterns caused by the synchronization jitter between the camera and the display. Then, DeepCCB-OCC incorporates convolutional neural network (CNN) models to accurately detect the pilot and data symbols in the transmission packets, regardless of the various D2C environments. Experiments with a commercial monitor and a smartphone demonstrate that DeepCCB-OCC outperforms the conventional CCB-OCC system from various distances and angles of a D2C link. The experiment results prove that, when the alignment angle was 20 degrees at a distance of 90 cm between the display and the camera, the proposed scheme achieved approximately 79.1 bps, which showed a performance improvement of 14.1% compared to the existing technique.

Deep neural networks (DNNs) have been used successfully for many image classification problems. One of the most important factors that determines the final efficiency of a DNN is the correct construction of the training set. Erroneously labeled training images can degrade the final accuracy and additionally lead to unpredictable model behavior, reducing reliability. In this paper, we propose MultiNET, a novel method for the automatic detection of noisy labels within image datasets. MultiNET is an adaptation of the current state-of-the-art confident learning method. In contrast to the original, our method aggregates the outputs of multiple DNNs and allows for the adjustment of detection sensitivity. We conduct an exhaustive evaluation, incorporating four widely used datasets (CIFAR10, CIFAR100, MNIST, and GTSRB), eight state-of-the-art DNN architectures, and a variety of noise scenarios. Our results demonstrate that MultiNET significantly outperforms the confident learning method.

In this paper, we present FVR-Net, which is a novel finger vein recognition network using a convolutional neural network (CNN) with a hybrid pooling mechanism. The scheme is based on the use of a block-wise feature extraction network to extract discrete features from interclass vein image samples, regardless of their visual quality. Input images to FVR-Net are subjected to preprocessing prior to being fed into the network in order to segment the vein patterns from the background. We designed a feature extraction network in which each block consists of a convolutional layer followed by hybrid pooling, whose output activation maps are concatenated before passing features to another block within the network. In the hybrid pooling layer, two subsampling layers of maxpooling and average pooling are placed in parallel where the former activates the most discrete features of the input, and the latter considers the complete extent of the input volume so better localization of features can be accessed. After the features are extracted, they are passed to three fully connected layers (FCLs) for classification. We conduct several experiments on two publicly available finger vein datasets based on visual quality of the images. When compared to conventional studies, the proposed model achieves outstanding recognition performance with accuracies of up to 97.84% and 97.22% for good and poor-quality images, respectively. By varying multiple network hyperparameters, we obtain optimal settings such that the model can guarantee the best recognition accuracy for a finger vein biometric system.

Analysis of reports published by the leading national centers for monitoring wildfires and other emergencies revealed that the devastation caused by wildfires has increased by 2.96-fold when compared to a decade earlier. The reports show that the total number of wildfires is declining; however, their impact on the wildlife appears to be more devastating. In recent years, deep neural network models have demonstrated state-of-the-art accuracy on many computer vision tasks. In this paper, we describe the design and implementation of a lightweight wildfire image classification model (LW-FIRE) based on convolutional neural networks. We explore different ways of using the existing dataset to efficiently train a deep convolutional neural network. We also propose a new method for dataset transformation to increase the number of samples in the dataset and improve the accuracy and generalization of the deep learning model. Experimental results show that the proposed model outperforms the state-of-the-art methods, and is suitable for real-time classification of wildfire images.

Object recognition is a fundamental concept in computer vision. Object detection models have recently played a vital role in various applications, including real-time and safety-critical systems such as camera surveillance and self-driving cars. Scientific research has proven that object detection models are prone to adversarial attacks. Although several proposed methods exist throughout the literature, they either target white-box models or specific-task black-box models and do not generalize on other detectors. In this paper, we proposed a new adversarial attack against Blackbox-based object detectors called AT-BOD. The proposed AT-BOD model can fool the single-stage and multi-stage detectors, where we used an optimization algorithm to generate adversarial examples depending only on the detector predictions. AT-BOD model works in two diverse ways, reducing the confidence score and misleading the model to make the wrong decision or hide the object detection models. Our solution achieved a fooling rate of 97% and a false negative increase of 99% on the YOLOv3 detector, and a fooling rate of 61% false-negative increase of 57% on the Faster R-CNN detector. The detection accuracy of YOLOv3 and Faster R-CNN under AT-BOD was dramatically reduced and reached ≤1% and ≤3%, respectively.

The ability to reliably detect and characterise defects embedded in austenitic steel welds depends on prior knowledge of microstructural descriptors, such as the orientations of the weld’s locally anisotropic grain structure. These orientations are usually unknown but it has been shown recently that they can be estimated from ultrasonic scattered wave data. However, conventional algorithms used for solving this inverse problem incur a significant computational cost. In this paper, we propose a framework which uses deep neural networks (DNNs) to reconstruct crystallographic orientations in a welded material from ultrasonic travel time data, in real-time. Acquiring the large amount of training data required for DNNs experimentally is practically infeasible for this problem, therefore a model based training approach is investigated instead, where a simple and efficient analytical method for modelling ultrasonic wave travel times through given weld geometries is implemented. The proposed method is validated by testing the trained networks on data arising from sophisticated finite element simulations of wave propagation through weld microstructures. The trained deep neural network predicts grain orientations to within 3° and in near real-time (0.04 s), presenting a significant step towards realising real-time, accurate characterisation of weld microstructures from ultrasonic non-destructive measurements. The subsequent improvement in defect imaging is then demonstrated via use of the DNN predicted crystallographic orientations to correct the delay laws on which the total focusing method imaging algorithm is based. An improvement of up to 5.3 dB in the signal-to-noise ratio is achieved.

It has been shown that the learning rate is one of the most critical hyper-parameters for the overall performance of deep neural networks. In this paper, we propose a new method for setting the global learning rate, named random amplify learning rates (RALR), to improve the performance of any optimizer in training deep neural networks. Instead of monotonically decreasing the learning rate, we expect to escape saddle points or local minima by amplifying the learning rate between reasonable boundary values based on a given probability. Training with RALR rather than conventionally decreasing the learning rate achieves further improvement on networks’ performance without extra consumption. Remarkably, the RALR is complementary with state-of-the-art data augmentation and regularization methods. Besides, we empirically study its performance on image classification tasks, fine-grained classification tasks, object detection tasks, and machine translation tasks. Experiments demonstrate that RALR can bring a notable improvement while preventing overfitting when training deep neural networks. For example, the classification accuracy of ResNet-110 trained on the CIFAR-100 dataset using RALR achieves a 1.34% gain compared with ResNet-110 trained traditionally.

Despite deep neural networks (DNNs) having achieved impressive performance in various domains, it has been revealed that DNNs are vulnerable in the face of adversarial examples, which are maliciously crafted by adding human-imperceptible perturbations to an original sample to cause the wrong output by the DNNs. Encouraged by numerous researches on adversarial examples for computer vision, there has been growing interest in designing adversarial attacks for Natural Language Processing (NLP) tasks. However, the adversarial attacking for NLP is challenging because text is discrete data and a small perturbation can bring a notable shift to the original input. In this paper, we propose a novel method, based on conditional BERT sampling with multiple standards, for generating universal adversarial perturbations: input-agnostic of words that can be concatenated to any input in order to produce a specific prediction. Our universal adversarial attack can create an appearance closer to natural phrases and yet fool sentiment classifiers when added to benign inputs. Based on automatic detection metrics and human evaluations, the adversarial attack we developed dramatically reduces the accuracy of the model on classification tasks, and the trigger is less easily distinguished from natural text. Experimental results demonstrate that our method crafts more high-quality adversarial examples as compared to baseline methods. Further experiments show that our method has high transferability. Our goal is to prove that adversarial attacks are more difficult to detect than previously thought and enable appropriate defenses.

Deep learning models have been widely used in natural language processing tasks, yet researchers have recently proposed several methods to fool the state-of-the-art neural network models. Among these methods, word importance ranking is an essential part that generates text adversarial examples, but suffers from low efficiency for practical attacks. To address this issue, we aim to improve the efficiency of word importance ranking, making steps towards realistic text adversarial attacks. In this paper, we propose CRank, a black box method utilized by our innovated masking and ranking strategy. CRank improves efficiency by 75% at the ’cost’ of only a 1% drop of the success rate when compared to the classic method. Moreover, we explore a new greedy search strategy and Unicode perturbation methods.

Over the recent years, Industry 4.0 (I4.0) technologies such as the Industrial Internet of Things (IIoT), Artificial Intelligence (AI), and the presence of Industrial Big Data (IBD) have helped achieve intelligent Fault Detection (FD) in manufacturing. Notably, data-driven approaches in FD apply Deep Learning (DL) techniques to help generate insights required for monitoring complex manufacturing processes. However, due to the ratio of instances where actual faults occur, FD datasets tend to be imbalanced, leading to training challenges that result in inefficient DL-based FD models. In this paper, we propose Dual Logits Weights Perturbation (DLWP) loss, a method featuring weight vectors for improved dataset generalization in FD systems. The weight vectors act as hyperparameters adjusted on a case-by-case basis to regulate focus accorded to individual minority classes during training. In particular, our proposed method is suitable for imbalanced datasets from safety-related FD tasks as it generates DL models that minimize false negatives. Subsequently, we integrate human experts into the workflow as a strategy to help safeguard the system. A subset of the results, model predictions with uncertainties exceeding a preset threshold, are considered a preliminary output subject to cross-checking by human experts. We demonstrate that DLWP achieves improved Recall, AUC, F1 scores.