Search

Deep learning (DL) libraries reduce the barriers to the DL model construction. In DL libraries, various building blocks are DL operators with different functionality, responsible for processing high-dimensional tensors during training and inference. Thus, the quality of operators could directly impact the quality of models. However, existing DL testing techniques mainly focus on robustness testing of trained neural network models and cannot locate DL operators’ defects. The insufficient test input and undetermined test output in operator testing have become challenging for DL library developers. In this article, we propose an approach, namely Duo, which combines fuzzing techniques and differential testing techniques to generate input and evaluate corresponding output. It implements mutation-based fuzzing to produce tensor inputs by employing nine mutation operators derived from genetic algorithms and differential testing to evaluate outputs’ correctness from multiple operator instances. Duo is implemented in a tool and used to evaluate seven operators from TensorFlow, PyTorch, MNN, and MXNet in an experiment. The result shows that Duo can expose defects of DL operators and realize multidimension evaluation for DL operators from different DL libraries.

Backfill coal mining, an environmentally friendly practice, is widely used to mitigate the hazards of water and sand inrushes (WSIs) and surface subsidence. In this study, a quantitative analysis for assessing the hazard reduction of WSIs due to backfilling was established that combines a modified analytic hierarchy process (AHP) with a geographic information system (GIS) and entropy. The analytical and quantitative model consists of four criteria and eight factors for the hazard assessment target layer. The weight of each index is comprehensively determined based on the modified AHP and entropy. Then, a hazard zone map that is color-coded is overlay analyzed based on GIS to comparatively evaluate the degrees of hazard from WSI due to longwall caving and paste backfill. The Taiping coal mine was used as a case study to validate the accuracy of the evaluation model. The results indicate that paste backfilling can effectively reduce the influence of the overburden failure height on WSI hazards. The highest hazard index of backfill mining was significantly reduced, and high-hazard and very high-hazard zones of longwall caving were transformed into low-hazard or even no-hazard zones. The evaluation results are in good agreement with actual engineering practices and offer an effective reference for practical engineering projects and prevention and control measures for safe mining under loose sand aquifers.

Backfill mining is an environmentally friendly and sustainable application that mitigates subsidence and prevents water inrush. A quantitative analysis method to assess the risks of water inrush and subsidence due to backfill mining was established combining a modified analytic hierarchy process (AHP) with fuzzy evaluation. The model considers three criteria and 12 factors to determine the weight of each index and criterion based on the modified AHP. Then, the fuzzy relation of the evaluation index to the risk degree was assessed. Following the fuzzy operation involving evaluation matrices, the risk degree was comprehensively evaluated based on a case study of the Lvgou Coal Mine. Mining activity was the most important factor in the risk of water inrush, followed by geological conditions. High-risk areas were transformed into low-risk areas by the use of backfill mining rather than caving mining, which effectively validated the accuracy of the evaluation model. This study also offers an effective reference source for practical engineering and for the formulation of prevention and control measures to ensure safe mining under aquifers and buildings.

Previous research has documented that political information in the mass media can shape attitudes and behaviors beyond voter choice and election turnout. The current study extends this body of work to examine associations between televised political campaign advertising (one of the most common forms of political communication people encounter) and worry about crime and violence in the context of the 2016 U.S. presidential election. We merge two large datasets—Kantar/CMAG data on televised campaign advertisement airings ( n = 3,767,477) and Simmons National Consumer Survey (NCS) data on television viewing patterns and public attitudes ( n = 26,703 respondents in the United States)—to test associations between estimated exposure to campaign ads about crime and crime worry, controlling for demographics, local crime rates, and political factors. Results from multivariate models show that estimated cumulative exposure to campaign ads about crime is associated with higher levels of crime worry. Exposure to campaign ads about crime increased crime worry among Republicans, but not Democrats.

In this article, the single-event response of the 1.2-kV silicon-carbide (SiC) power MOSFETs with varied buffer layer designs is investigated by the 2-D numerical simulations. The structural parameters of the buffer layers are compared and analyzed to understand the transient response after the heavy ion strike and the related physical mechanisms comprehensively. Simulation results reveal that an optimized single buffer structure can be acquired by using a relatively thicker buffer layer (T $\mu \text{m}$ ) and a moderate doping concentration (D cm−3). It demonstrates that the single-event-burnout (SEB) performance can be improved significantly under the worst case bias conditions [a drain bias voltage of 1.2 kV and a linear energy transfer (LET) of 1 pC/ $\mu \text{m}$ ]. In addition, the optimized structural combinations adopted in the dual or triple buffer layers can strengthen the SEB performance further. The simulation results show that a step electric field distribution is established inside the optimized dual or triple buffers, where the electric field peak is mitigated from 3 to 2.2 MV/cm. Moreover, the excess carrier generation is suppressed and the local temperature rise is weakened during the transient electrothermal process. Consequently, the structure with the optimal buffer layer designs can enlarge the SEB safe operating area (SOA) from 30%–50% to 100% rated breakdown voltage at high LET bias, which makes the SiC power MOSFETs used in aerospace and aviation power electronic systems become possible.

A unique three-step short-circuit protection method is proposed for the 650-V enhancement mode (E-mode) gallium nitride high-electron mobility transistor (GaN HEMT). This method can quickly detect the short-circuit event, reduce gate voltage to enhance the device short-circuit capability, and turn off the device under fault after confirmation. Experimental results prove that with this method, the short-circuit fault detection time for E-mode GaN HEMT is shortened from 2 μ s to several tens of nanoseconds, and the device can be successfully protected from fatal failure under high dc bus voltage without mistriggering.

PurposeThe high uncertainty of technological innovation in megaprojects brings great challenges to the R&D institution and also acts as a trigger for moral hazard. The incentive and supervision are effective means to improve the performance of innovation. The purpose of this paper is to propose appropriate incentive and supervision mechanisms to reduce information asymmetry and improve the efficiency of incentives. Suggestions on technological innovation are put forward to megaprojects management.Design/methodology/approachAccording to the principal-agent theory, the research develops incentive models under three states, i.e. information symmetry, information asymmetry and information asymmetry based on supervision mechanism. The Bayesian theory is employed to prove the effectiveness of the novel supervision method based on risk assessment.FindingsThe results indicate that under the information asymmetry, the incentive intensity is positively correlated with the social benefits coefficient, and negatively correlated with the patent benefits coefficient. The R&D effort and the owner's incentive intensity decline with the increase of information asymmetry. The supervision of risks can effectively reduce the degree of information asymmetry, and the higher the uncertainty of innovations, the more significant the effect of supervision is. As the supervision intensity increases, the incentive intensity, the R&D effort and the innovation output will increase. In addition, the R&D institutions with high innovation capability, low unit cost of R&D and low risk-aversion are more willing to make efforts to innovate.Originality/valueThis study fills the research gap on incentive and supervision of technological innovation in megaprojects. The externality of innovation benefits is considered in the model. The traditional incentive model is extended through the introduction of supervision. Furthermore, a novel supervision method based on risk assessment is proposed. The results validate the importance of risk management in technological innovation and provide a new insight for project management.

The newly emerging task of person search with natural language query aims at retrieving the target pedestrian by a text description of the pedestrian. It is more applicable compared to person search with image/video query, i.e., person re-identification. In this paper, we propose a novel Adversarial Attribute-Text Embedding (AATE) network for person search with text query. In particular, a cross-modal adversarial learning module is proposed to learn discriminative and modality-invariant visual-textual features. It consists of a cross-modal learner and a modality discriminator, playing a min-max game in an adversarial learning way. The former is to improve intra-modality discrimination and inter-modality invariance towards confusing the modality discriminator. The latter is to distinguish the features from different modalities and boost the learning of modality-invariant features. Moreover, a visual attribute graph convolutional network is proposed to learn visual attributes of pedestrians, which possess better descriptiveness, interpretability and robustness compared to pedestrian appearance features. A hierarchical text embedding network, consisting of multi-stacked bidirectional LSTMs and a textual attention block, is developed to extract effective textual features from text descriptions of pedestrians. Extensive experimental results on two challenging benchmarks, have demonstrated the effectiveness of the proposed approach.

Cancer information seeking and scanning predict a variety of preventive health behaviors. However, previous work has rarely gauged seeking and scanning of specific cancer screening information. Moreover, colorectal cancer prevalence and mortality rates are higher among black than white Americans and it remains unclear if these groups differ in their cancer screening information acquisition patterns. We surveyed black and white Americans between 45 and 74 years of age to investigate rates, sources, and correlates of colorectal cancer screening (CRCS) information seeking and scanning. Black and white Americans had similar likelihoods of engaging in information seeking and scanning regarding CRCS. However, black Americans reported using significantly more sources for CRCS information seeking and scanning than did white Americans. Both screening test-specific information seeking and scanning are associated with stool-based tests, but only information seeking is associated with flexible sigmoidoscopy or colonoscopy. We discuss study implications for reaching out to different racial groups to promote colorectal cancer screening behavior.

In order to reduce the operating cost of hydrogen fuel cell hybrid electric tram, this paper proposes a hierarchical control method, which includes a control layer and a management layer. In the control layer, through three-dimensional efficiency, modeling, and geometric solution of fuel cell system (FCS), the equivalent fitting circle method is adopted to realize the optimal allocation among sets of multiple fuel cells with different parameters. Meanwhile, in management layer, according to the operation states of tram, the supercapacitor dynamic energy consumption coefficient is proposed and applies in the tram efficiency model for the equivalent energy consumption (EEC) method, and the improved EEC method can accurately provide the optimal power distribution of the hybrid system under different operating conditions. According to RT-LAB platform tests, compared with other online methods under the same operating conditions, the proposed method has the apparent satisfactory results in reducing hydrogen consumption. What is more, the actual tram operation test also shows that the proposed method ensures the output stability of the FCS so as to prolong its life and reduce its operation and maintenance costs.

From the perspective of non-financial performance, this study investigates the effect of corporate awards on executive compensation. We find that as a measure of non-financial performance, corporat...

This paper at first presents the proposal and design of a filtering crossover and diplexer with two second-order filtering channels on substrate integrated waveguide (SIW) quadruple-mode resonators (QMRs) and in single cavities. The proposed method has an interesting application in the overall size miniaturization design. The proposed resonator is yielded with four perturbed modes (TE103, TE104, TE201 and TE202) featuring distinct electric field distributions by inserting metal via-holes in the center of a rectangular SIW cavity. Next, a pair of resonances is applied to form a second-order filter channel while TE201 and TE202 modes are implemented to realize the other, thereby integrating the whole circuit in a single cavity. Besides, the transmission responses between the channels are isolated only by the perturbed via-holes. Slots are introduced in the metal surface to further increase each channel design flexibility. In this way, controllable frequencies and compact size have been well achieved. Two examples including a crossover and diplexer with two second-order filter channels are designed, fabricated, and measured to verify. Good agreement between simulation and measurement can be observed.

Mucin 13 (MUC13) is a glycoprotein that is expressed on the cell surface and participates in the tumorigenesis of multiple malignancies, including pancreatic cancer, colorectal cancer and renal cancer. However, to the best of our knowledge, the expression levels and function of MUC13 in lung cancer progression have not yet been demonstrated. Therefore, the present study examined the expression pattern and regulatory role of MUC13 in lung cancer tumorigenesis. The results demonstrated that MUC13 was highly expressed in lung cancer tissues and cell lines compared with that in normal tissues and cell lines. Functionally, knockdown of MUC13 inhibited cell proliferation and enhanced the apoptosis of A549 and NCI-H1650 lung cancer cells. Furthermore, silencing of MUC13 suppressed the migration and invasion of lung cancer cells. Additionally, a xenograft tumor model demonstrated that knockdown of MUC13 delayed the development of the lung cancer xenograft and suppressed the expression of proliferation marker Ki-67 in tumor tissues. Mechanistically, MUC13 activated the ERK signaling pathway by enhancing the phosphorylation of ERK, JNK and p38 in lung cancer tissues compared with that in normal tissues. Knockdown of MUC13 inhibited the phosphorylation of ERK/JNK/p38 in A549 and NCI-H1650 cells. Overall, these findings suggested that MUC13 could act as an oncogenic glycoprotein to accelerate the progression of lung cancer via abnormal activation of the ERK/JNK/p38 signaling pathway and might serve as a therapeutic target for lung cancer treatment.

Active learning has recently attracted increasing attention in the task of person re-identification, due to its unique scalability that not only maximally reduces the annotation cost but also retains the satisfying performance. Although some preliminary active learning methods have been explored in scalable person re-identification task, they have the following two problems: 1) the inefficiency in the selection process of image pairs due to the huge search space, and 2) the ineffectiveness caused by ignoring the impact of unlabeled data in model training. Considering that, we propose a Multi-grained Active Semi-Supervised learning framework, named MASS, to address the scalable person re-identification problem existing in the practical scenarios. Specifically, we firstly design a cluster-scatter procedure to alleviate the inefficiency problem, which consists of two components: cluster step and scatter step. The cluster step shrinks the search space into individual small clusters by a coarse-grained clustering method, and the subsequent scatter step further mines the hard distinguished image pairs from unlabelled set to purify the learned clusters by a novel centrality-based adaptive purification strategy. Afterward, we introduce a customized purification loss for the purified clustering, which utilizes the complementary information in both labeled and unlabeled data to optimize the model for solving the ineffectiveness problem. The cluster-scatter procedure and the model optimization are performed in an iterative fashion to achieve the promising performance while greatly reducing the annotation cost. Extensive experimental results have demonstrated that MASS can even achieve a competitive performance with fully supervised methods in the case of extremely less annotation requirements.

The quality of video representation directly decides the performance of video related tasks, for both understanding and generation. In this paper, we propose single-modality pretrained feature fusion technique which is composed of reasonable multi-view feature extraction method and designed multi-modality feature fusion strategy. We conduct comprehensive ablation studies on MSR-VTT dataset to demonstrate the effectiveness of proposed method and it surpasses the state-of-the-art methods on both MSR-VTT and VATEX datasets. We further propose the multi-modality pretrained model finetuning technique and dataset augmentation scheme to improve the model's generalization capability. Based on these two proposed pretraining techniques and dataset augmentation scheme, we win the first place in the video captioning track of the MM21 pretraining for video understanding challenge.

Compared with expensive pixel-wise annotations, image-level labels make it possible to learn semantic segmentation in a weakly-supervised manner. Within this pipeline, the class activation map (CAM) is obtained and further processed to serve as a pseudo label to train the semantic segmentation model in a fully-supervised manner. In this paper, we argue that the lost structure information in CAM limits its application in downstream semantic segmentation, leading to deteriorated predictions. Furthermore, the inconsistent class activation scores inside the same object contradicts the common sense that each region of the same object should belong to the same semantic category. To produce sharp prediction with structure information, we introduce an auxiliary semantic boundary detection module, which penalizes the deteriorated predictions. Furthermore, we adopt smoothness loss to encourage prediction inside the object to be consistent. Experimental results on the PASCAL-VOC dataset illustrate the effectiveness of the proposed solution.

Deep learning(DL) techniques attract people from various fields with superior performance in making progressive breakthroughs. To ensure the quality of DL techniques, researchers have been working on testing and verification approaches. Some recent studies reveal that the underlying DL operators could cause defects inside a DL model. DL operators work as fundamental components in DL libraries. Library developers still work on practical approaches to ensure the quality of operators they provide. However, the variety of DL operators and the implementation complexity make it challenging to evaluate their quality. Operator testing with limited test cases may fail to reveal hidden defects inside the implementation. Besides, the existing model-to-library testing approach requires extra labor and time cost to identify and locate errors, i.e., developers can only react to the exposed defects. This paper proposes a fuzzing-based operator-level precision testing approach to estimate individual DL operators' precision errors to bridge this gap. Unlike conventional fuzzing techniques, valid shape variable inputs and fine-grained precision error evaluation are implemented. The testing of DL operators is treated as a searching problem to maximize output precision errors. We implement our approach in a tool named Predoo and conduct an experiment on seven DL operators from TensorFlow. The experiment result shows that Predoo can trigger larger precision errors compared to the error threshold declared in the testing scripts from the TensorFlow repository.

Deep learning has made great progress in medical diagnosis. However, due to data standardization and privacy restriction, the acquisition and sharing of medical image data have been hindered, leading to the unacceptable accuracy of some intelligent medical diagnosis models. Another concern is data quality. If insufficient quantity and low-quality data are used for training and testing medical diagnosis models, it may cause serious medical accidents. We always use data augmentation to deal with it, and one of the most representative ways is through mutation relation. However, although common mutation methods can increase the amount of medical data, the quality of the image cannot be guaranteed due to the particularity of medical image. Therefore, combined with the characteristics of medical images, we propose TauMed, which implements augmentation techniques based on a series of mutation rules and domain semantics on medical datasets to generate sufficient and high-quality images. Moreover, we chose the ResNet-50 model to experiment with the augmented dataset and compared the results with two main popular mutation tools. The experimental result indicates that TauMed can improve the classification accuracy of the model effectively, and the quality of augmented images is higher than the other two tools. Its video is at https://www.youtube.com/watch?v=O8W8I7U_eqk and TauMed can be used at http://121.196.124.158:9500/.

In this Letter, a 1.2 kV SiC power MOSFET with a partially widened retrograde P-well (RP) structure and N-implanting region is proposed to enhance the device's reliability. Compared with the conventional SiC power MOSFET, the short circuit (SC) ability of the proposed structure can be improved effectively without sacrificing other performance. Simulation results reveal that a 40% reduction of the SC saturation current can be achieved, resulting in the SC withstand time increase from 7 to 10 μs at the DC-link voltage 800 V. Moreover, a better gate oxide reliability at 1.2 kV blocking condition also can be achieved. The peak electric field at the gate oxide interface is decreased by ∼48% owing to the shielding effect of the RP structure. In addition, the fabrication technology of the proposed structure is compatible with the standard planar SiC MOSFET manufacture process only with a few additional implanting steps. Therefore, this new MOSFET structure provides a simple and effective way to optimise the reliability of the planar SiC power MOSFET.

A buried gate fin p-body insulated gate bipolar transistor (BG-Fin-P IGBT) is proposed to achieve ultralow on-state voltage drop (V CE(sat)) and good short-circuit (SC) ruggedness simultaneously. A buried gate is introduced at the bottom part of the fin structure, forming a local region with the nanoscale mesa width, which enhances the conductivity modulation effectively. Meanwhile, a relatively wide mesa width (>0.5 μm) can be adopted at the main fin structure to maintain a good SC capability. Compared to the previously reported ultra-narrow-mesas fin p-body IGBT, simulation results reveal that the V CE(sat) of the BG-Fin-P IGBT is reduced from 1.39 to 1.03 V at the current density of 100 A/cm2 without SC ability degradation. Meanwhile, more than 10 μs short circuit withstand time is enabled at the junction temperature of 423 K for all structures. Moreover, the proposed structure can avoid a fabrication difficulty of the emitter contact when a very narrow mesa width (∼30 nm) is required to achieve the ultralow V CE(sat), which brings design freedom on the device's structure.

In the cause of working out the challenge of remaining life prediction (RUL) of proton exchange membrane fuel cell (PEMFC) under dynamic operating conditions, this article proposes a PEMFC RUL forecast technique based on the sparse autoencoder (SAE) and deep neural network (DNN). The method extracts the data set from the original experimental data at intervals periods of one hour to realize datum reconstruction. The Gaussian-weighted moving average filter is used to smooth noisy data (voltage and current). The smoothed filtered power output signal of the stack is extracted as an aging indicator. The SAE is used to extract the prediction features automatically, and the DNN is applied to realize the RUL prediction. The proposed method is experimentally verified using 127 369 experimental data. The effectiveness of the novel method is verified by three different training sets and test set configurations. The experimental results reveal that the novel approach has the best prediction effectiveness when the training set length is set to 500 h. At this point, the prediction accuracy can reach 99.68%. The mean absolute error (MAE), mean square error (MSE), and root-mean-square error (RMSE) are minimum values, which are 0.2035, 0.1121, and 0.3348, respectively. The superiority and effectiveness of the proposed approach are further validated by comparison with the K-nearest neighbor and support vector regression machine. The proposed approach can be appropriate for the prediction of the RUL of PEMFC under dynamic conditions.

Jiawei Liu,1,* Yan Xu,1,* Muxin Yu,1,* Zhao Liu,1,2,* Yi Xu,3,* Ge Ma,1 Wenbin Zhou,1 Peng Kong,1 Lijun Ling,1 Shui Wang,1 Hong Pan,1 Yi Zhao1 1Department of Breast Surgery, The First Affiliated Hospital with Nanjing Medical University, Nanjing 210029, People’s Republic of China; 2Department of Thyroid and Breast Surgery, Affiliated Hospital of Xuzhou Medical University, Xuzhou 221000, People’s Republic of China; 3Department of Pathology, The First Affiliated Hospital with Nanjing Medical University, Nanjing 210029, People’s Republic of China*These authors contributed equally to this workCorrespondence: Hong Pan; Yi ZhaoDepartment of Breast Surgery, The First Affiliated Hospital with Nanjing Medical University, 300 Guangzhou Road, Nanjing 210029, People’s Republic of ChinaTel/Fax +86-25-83718836Email pamghong@163.com; doctorzhaoyi@sina.cnBackground: Circulating tumor cells (CTCs) intravasate into the bloodstream throughout early cancer stages, promoting metastasis. The tumor microenvironment plays a crucial role in disease progression and outcome. The aim of this prospective study was to investigate the associations of intratumoral and stromal tumor-infiltrating lymphocytes (TILs) with CTCs among patients receiving neoadjuvant chemotherapy (NAC).Methods: We analyzed CTCs in 30 patients with primary breast cancer before and after NAC. The numbers of intratumoral TILs (iTILs) and stromal TILs (sTILs) from pre-NAC formalin-fixed paraffin-embedded core biopsies and post-NAC surgical samples were analyzed. The associations of TILs with pathologic complete response (pCR) and outcome were also evaluated.Results: Of the 30 patients, pCR was achieved in nine (30.0%) patients. A total of 25 (83.3%) patients were CTC-positive before NAC, and eight (26.7%) patients were CTC-positive after NAC. Neither CTC detection before NAC nor CTC after NAC was predictive of pCR. Nevertheless, the presence of CTCs after NAC was significantly associated with early metastatic relapse (P = 0.049) and worse disease-free survival (P = 0.009). After NAC, total sTILs, CD4+ T cells, and CD8+ T cells were significantly correlated with CTC detection. Increased infiltration of sTILs and CD4+ T cells was also an unfavorable prognostic factor as measured by the rate of metastatic relapse.Conclusion: Detection of CTCs after NAC was positively associated with the metastatic relapse of breast cancer patients. Increased infiltration of sTILs after NAC was correlated with CTCs and was found to be an unfavorable prognostic factor.Keywords: circulating tumor cells, tumor infiltrating lymphocytes, neoadjuvant chemotherapy, breast cancer, metastatic relapse

In this study, Big Data Analytics has been applied to implement smart manufacturing services performed by local commercial laundry Small and Medium Sized Enterprises (SMEs), which, to be sp...

Non-tariff measures (NTMs) such as technical barriers to trade (TBT) and sanitary and phytosanitary measures (SPMs) have become new trade barriers, which is contrary to these measures’ original int...

Studying the haemodynamics of the central shunt (CS) and modified Blalock–Taussig shunt (MBTS) benefits the improvement of postoperative recovery for patients with an aorta-pulmonary shunt. Shunt configurations, including CS and MBTS, are virtually reconstructed for infants A and B based on preoperative CT data, and three-dimensional models of A, 11 months after CS, and B, 8 months after MBTS, are reconstructed based on postoperative CT data. A series of parameters including energy loss, wall shear stress, and shunt ratio are computed from simulation to analyse the haemodynamics of CS and MBTS. Our results showed that the shunt ratio of the CS is approximately 30% higher than the MBTS and velocity distribution in the left pulmonary artery (LPA) and right pulmonary artery (RPA) was closer to a natural development in the CS than the MBTS. However, energy loss of the MBTS is lower, and the MBTS can provide more symmetric pulmonary artery (PA) flow than the CS. With the growth of infants A and B, the shunt ratio of infants was decreased, but maximum wall shear stress and the distribution region of high wall shear stress (WSS) were increased, which raises the probability of thrombosis. For infant A, the preoperative abnormal PA structure directly resulted in asymmetric growth of PA after operation, and the LPA/RPA ratio decreased from 0.49 to 0.25. Insufficient reserved length of the MBTS led to traction phenomena with the growth of infant B; on the one hand, it increased the eddy current, and on the other hand, it increased the flow resistance of anastomosis, promoting asymmetric PA flow.

Person re-identification aims at identifying a certain pedestrian across non-overlapping multi-camera networks in different time and places. Existing person re-identification approaches mainly focus on matching pedestrians on images; however, little attention has been paid to re-identify pedestrians in videos. Compared to images, video clips contain motion patterns of pedestrians, which is crucial to person re-identification. Moreover, consecutive video frames present pedestrian appearance with different body poses and from different viewpoints, providing valuable information toward addressing the challenge of pose variation, occlusion, and viewpoint change, and so on. In this article, we propose a Dense 3D-Convolutional Network (D3DNet) to jointly learn spatio-temporal and appearance representation for person re-identification in videos. The D3DNet consists of multiple three-dimensional (3D) dense blocks and transition layers. The 3D dense blocks enlarge the receptive fields of visual neurons in both spatial and temporal dimensions, leading to discriminative appearance representation as well as short-term and long-term motion patterns of pedestrians without the requirement of an additional motion estimation module. Moreover, we formulate a loss function consisting of an identification loss and a center loss to minimize intra-class variance and maximize inter-class variance simultaneously, toward addressing the challenge of large intra-class variance and small inter-class variance. Extensive experiments on two real-world video datasets of person identification, i.e., MARS and iLIDS-VID, have shown the effectiveness of the proposed approach.

In this paper, a novel WR-2.8 band (260-400 GHz) pseudo-elliptic waveguide bandpass filter (BPF) based on a singlet and an extracted pole resonator is proposed. The singlet based on TE301 mode and extracted pole resonant cavities is developed to generate two transmission zeros (TZs) on both sides of the passband in order to achieve high selectivity. The mechanism of TZs generation of both structures is analyzed in this paper, and TZs position is predicted precisely. Furthermore, each TZ position can be controlled independently by changing the dimensions of resonant cavities. The proposed filter fabricated by conventional computer numerical control (CNC) milling technology exhibits an insertion loss (IL) around 0.7dB, a 3-dB fractional bandwidth (FBW) of 9.9% centered at 357GHz and an ultra-high selectivity with a 0.87 30-dB rectangular factor which is all in good agreement with the simulations. To the best of the author's knowledge, 0.87 is the highest 30-dB rectangular factor among such wide fractional bandwidth THz BPF in the open literature. This high-performance filter is capable of improving the sideband rejection receiver performance.

Image inpainting technique recovers the missing regions of an image using information from known regions and it has shown success in various application fields. As a popular kind of methods, Markov Random Field (MRF)-based methods are able to produce better results than earlier diffusion-based and sparse-based methods on inpainting images with big holes. However, for images with complex structures, the results are still not quite pleasant and some inpainting trails exist. The direction feature is an important factor for image understanding and human eye visual requirements, and exploiting multi-direction features is of great potential to further improve inpainting performance. Following the idea, this paper proposes a Structure Offsets Statistics based image inpainting algorithm by exploiting multiple direction features under the framework of MRF-based methods. Specifically, when selecting proper labels, multi-direction features are extracted and applied to construct a structure image and a non-structure image, and the candidate labels are chosen from the offsets of structure and non-structure images. Meanwhile, the multi-direction features are applied to construct a new smooth term for the energy equation which is then solved by graph-cut optimization technology. Experimental results show that on inpainting tasks with various complexities, the proposed method is superior to several state-of-the-art approaches in terms of the abilities of maintaining structure coherence and neighborhood consistence and the computational efficiency.

For solving the problem of sequence failure diagnosis of proton exchange membrane fuel cell (PEMFC) water management subsystem, this paper proposes a PEMFC failure diagnosis method of time series based on the bidirectional long short-term memory (BiLSTM) network and t-distributed stochastic neighbor embedding (t-SNE). This approach adopts the normalization strategy to eliminate the influence caused by dimensional differences of different parameters. The t-SNE is presented to decrease the dimensionality of normalized data to the estimate of intrinsic dimensionality to extract key characteristic variables. The width of the diagnostic window is set to transform the original single moment diagnosis problem into the fault diagnosis problem of multi-variable time series, which is more consistent with the time scale and physical evolution law of the PEMFC water management fault generation. The 672 sets of training sets and 448 sets of test sets are learned and tested by the BiLSTM. The experimental results show that the BiLSTM-tSNE method can realize the sequence fault diagnosis of the PEMFC water management subsystem with 96.88% diagnostic accuracy and 24 s of operation time. Compared with the conventional approach of multi-class support vector machine algorithm, the training accuracy and the testing accuracy of the proposed method are improved by 15% and 16.88%, respectively. The operation time of the presented approach is only about 1/28 of the multi-class support vector machine algorithm.

More and more tenants (enterprises and personal users) migrate their tasks to clouds since it is a simple and low-cost way to obtain enough computing resources. However, due to potential node failures and malicious tenants, the modern cloud encounters one critical challenge, i.e., robustness. Conventionally, the cloud vendors deploy auxiliary systems to protect the cloud, which requires additional resource cost and increases the network complexity. To enhance the system robustness, this paper proposes a complementary scheme to improve the cloud robustness through efficient VM placement. Specifically, to alleviate the impact of malicious tenants and node failures on the cloud, when deploying VMs, we limit the number of pods (or service nodes) that each tenant can access, and the number of tenants hosted by each pod (or service node). Though there are a lot of works on VM placement, it is very challenging when the robustness issue is taken into consideration. To solve this problem, we formulate an integer linear programming and propose a rounding-based algorithm with a logarithmic approximation ratio. The simulation results show the high efficiency of the proposed algorithm. For example, our algorithm can improve the network throughput by 150% with other alternatives.

This paper investigates the short-circuit (SC) capability of the 3.3-kV 5-A silicon carbide (SiC) metal–oxide–semiconductor field-effect transistor (MOSFET) from GeneSiC (Generation-1, engineering sample). The SC withstand time (SCWT) of the tested 3.3-kV device could not reach the benchmark of 10-μs at a 2.2-kV bus voltage and 18-V gate voltage. A three-step ultra-fast SC protection method is introduced and validated. It can detect a SC fault and reduce the saturation current within 80 ns, then softly turn off the device within 2 μs. Using this protection method, the SC energy can be reduced by around 32%. Additionally, a noise immunity test showed this protection would not be falsely triggered at the device’s rated current. Medium-voltage (MV) SiC MOSFET based power conversion systems could utilize this method to enhance their SC capabilities without incurring efficiency losses.

Venipuncture is a common medical procedure. The use of augmented reality-based assistive devices can improve the first puncture success rate in patients with poor vascular filling. In order to improve the image rendering quality and speed of auxiliary equipment, this study develop a two-dimensional matched filtering algorithm on a Field Programmable Gate Array (FPGA) in a near-infrared vein imaging system, which use parallel processing to offer real-time response and is designed as a small handheld portable device. A customized dorsal hand vein image library with 200 images captured from 120 participants is used to analyze the effects of convolution kernel parameters and exposure time on vascular imaging with different depths, and the correlation model between these parameters and vascular depth are constructed. We use the Tenengrad, variance, Laplace smoothness and standard deviation as evaluation indicators, and compare our algorithm with three other related studies. Experimental results show that the rendering quality of our proposed algorithm is significantly higher than other algorithms. In addition, the rendering speed of our algorithm can reach 66 fps, which is twice faster than the current fastest algorithm.

Objective: To evaluate the efficacy and safety of anti-PD-1/PD-L1 Inhibitors versus docetaxel for non-small cell lung cancer by meta-analysis.Methods: Randomized controlled trials (RCTs) about anti-PD-1/PD-L1 Inhibitors versus docetaxel on the treatment of NSCLC were searched in CNKI, WF, VIP, PubMed, EMBASE, Cochrane Library, and Web of Science databases. Two reviewers independently screened literature, extracted data and evaluated the risk of bias of eligible studies. Meta-analysis was performed by RevMan5.3 software.Results: Compared with the use of docetaxel chemotherapy for NSCLC, the overall survival and progression-free survival of the anti-PD-1/PD-L1 Inhibitors regimen are better [overall survival: (HR= 0.73, 95%CI:0.69∼0.77, PConclusion: Compared with the docetaxel chemotherapy regimen, the anti-PD-1/PD-L1 Inhibitors has certain advantages in terms of efficacy and safety. The results still need to be confirmed by a multi-center, large sample, and high-quality research.

Semi-supervised learning on graphs is an important problem in the machine learning area. In recent years, state-of-the-art classification methods based on graph neural networks (GNNs) have shown their superiority over traditional ones such as label propagation. However, the sophisticated architectures of these neural models will lead to a complex prediction mechanism, which could not make full use of valuable prior knowledge lying in the data, e.g., structurally correlated nodes tend to have the same class. In this paper, we propose a framework based on knowledge distillation to address the above issues. Our framework extracts the knowledge of an arbitrary learned GNN model (teacher model), and injects it into a well-designed student model. The student model is built with two simple prediction mechanisms, i.e., label propagation and feature transformation, which naturally preserves structure-based and feature-based prior knowledge, respectively. In specific, we design the student model as a trainable combination of parameterized label propagation and feature transformation modules. As a result, the learned student can benefit from both prior knowledge and the knowledge in GNN teachers for more effective predictions. Moreover, the learned student model has a more interpretable prediction process than GNNs. We conduct experiments on five public benchmark datasets and employ seven GNN models including GCN, GAT, APPNP, SAGE, SGC, GCNII and GLP as the teacher models. Experimental results show that the learned student model can consistently outperform its corresponding teacher model by 1.4% - 4.7% on average. Code and data are available at https://github.com/BUPT-GAMMA/CPF

The evolution of fractures in overburden is quantitatively investigated to characterize the effect of mining activity. Scale model testing and numerical modeling were used based on the engineering geological and mining environments of Panel 11050 in the Quandian Coalmine in Henan Province, China. The maximum vertical displacement is 62.76 m, which is 140 m from the initial mining in the scale model test. Based on the fractal geometric theory, the fractal dimensions of the fractures in the overburden are calculated and visualized. The results reveal that if two coal seams are mined at the same time, the fractal dimension of the fracture network in the overburden increase with the progression of mining, but the rate of increase gradually slows. The relationships between the fractal dimension and the maximum height of the overburden failure and maximum overburden subsidence are nonlinear. The structural characteristics of the overburden are represented by the network of fractures. With increasing distance from the coal seam roof, the mining stress is gradually transferred upward to the overlying strata, and the scale of this stress transfer gradually reduces. The variation in the vertical stress gradually weakens and shows a delayed change with the mining process. The maximum principal stress is compressive stress and is distributed in an “arch” shape. The stress on both sides of the “arch” is high, and the intermediate stress is low. The stress within the “arch” shows an opposite trend.

The Shanghai East China Electric Power Building, which was completed in 1988, is widely accepted as one of the first postmodern high-rise buildings in Shanghai. Based on articles published in mass media and professional magazines, interviews with relevant stakeholders and social media debates, this paper focuses on two controversies regarding the building’s peculiar architectural form. The first occurred between 1988 and 1992, when the building’s postmodern appearance aroused heated debates among architectural professionals. The second happened between 2015 and 2018, when the building’s postmodern appearance was planned to be replaced with a slated Art Deco surface during its renovation into a boutique hotel. This paper reveals how a thirst for ‘form innovation’ emerged in the specific social and professional environment shortly after China’s opening-up, and how professional and public awareness of the value of late 20th century architectural heritage was stimulated in the early 21st century in the search for an alternative representation of urban identity other than the widely accepted Art Deco style. This paper emphasises the public meaning of architectural forms in arguing for institutional co-operation in systematic evaluation and conservation legislation for late twentieth century Chinese architectural heritage sites to maintain the historical diversity of the cityscape during urban regeneration.

Osteoarthritis (OA) is one of the most frequent musculoskeletal diseases characterized by degeneration of articular cartilage, subchondral bone remodeling, and synovial membrane inflammation, which is a leading cause of global disability, morbidity, and decreased quality of life. Interpreting the potential mechanisms of OA pathogenesis is essential for developing novel prevention and disease-modifying therapeutic interventions. Gut microbiota is responsible for a series of metabolic, immunological, and structural and neurological functions, potentially elucidating the heterogeneity of OA phenotypes and individual features. In this narrative review, we summarized research evidence supporting the hypothesis of a “gut-joint axis” and the interaction between gut microbiota and the OA-relevant factors, including age, gender, genetics, metabolism, central nervous system, and joint injury, elucidating the underlying mechanisms of this intricate interaction. In the context, we also speculated the promising manipulation of gut microbiota in OA management, such as exercise and fecal microbiota transplantation (FMT), highlighting the clinical values of gut microbiota. Additionally, future research directions, such as more convincing studies by the interventions of gut microbiota, the gene regulation of host contributing to or attributed to the specific phenotypes of gut microbiota related to OA, and the relevance of distinct cell subgroups to gut microbiota, are expected. Moreover, gut microbiota is also the potential biomarker related to inflammation and gut dysbiosis that is able to predict OA progression and monitor the efficacy of therapeutic intervention.

This paper presents our endeavor for solving task11, NLPContributionGraph, of SemEval-2021. The purpose of the task was to extract triples from a paper in the Nature Language Processing field for constructing an Open Research Knowledge Graph. The task includes three sub-tasks: detecting the contribution sentences in papers, identifying scientific terms and predicate phrases from the contribution sentences; and inferring triples in the form of (subject, predicate, object) as statements for Knowledge Graph building. In this paper, we apply an ensemble of various fine-tuned pre-trained language models (PLM) for tasks one and two. In addition, self-training methods are adopted for tackling the shortage of annotated data. For the third task, rather than using classic neural open information extraction (OIE) architectures, we generate potential triples via manually designed rules and develop a binary classifier to differentiate positive ones from others. The quantitative results show that we obtain the 4th, 2nd, and 2nd rank in three evaluation phases.

With the change of people’s consumption concept, customers not only consider the characteristics of the product itself, but also gradually consider the brand. At present, the competition in the product market is not only the competition of the product itself, but also the competition between different brands. As a marketing method, brand stories are gradually being widely used by major companies, which creates product differences to a certain extent. In order to explore the influence of brand stories on customers’ purchase intention, this paper takes the social value, emotional value and cognitive value of brand stories from three dimensions, and introduces brand identity as an intermediary variable to construct a research model of brand stories on customers’ purchase intention. The empirical results show that: social value, emotional value and cognitive value have a significant positive impact on brand identity; brand identity has a significant positive impact on customer purchase intention; social value, emotional value and cognitive value have a direct impact on customer purchase intention. Therefore, the study of brand stories from multiple perspectives enables enterprises to have a deep understanding of consumers’ purchase intention and has certain guiding significance for the future development of enterprises.

Occluded person re-identification (ReID) aims to match person images with occlusion. It is fundamentally challenging because of the serious occlusion which aggravates the misalignment problem between images. At the cost of incorporating a pose estimator, many works introduce pose information to alleviate the misalignment in both training and testing. To achieve high accuracy while preserving low inference complexity, we propose a network named Pose-Guided Feature Learning with Knowledge Distillation (PGFL-KD), where the pose information is exploited to regularize the learning of semantics aligned features but is discarded in testing. PGFL-KD consists of a main branch (MB), and two pose-guided branches, \ieno, a foreground-enhanced branch (FEB), and a body part semantics aligned branch (SAB). The FEB intends to emphasise the features of visible body parts while excluding the interference of obstructions and background (\ieno, foreground feature alignment). The SAB encourages different channel groups to focus on different body parts to have body part semantics aligned representation. To get rid of the dependency on pose information when testing, we regularize the MB to learn the merits of the FEB and SAB through knowledge distillation and interaction-based training. Extensive experiments on occluded, partial, and holistic ReID tasks show the effectiveness of our proposed network., Comment: ACM MM 2021

Health information sharing has become especially important during the COVID-19 (coronavirus disease 2019) pandemic because people need to learn about the disease and then act accordingly. This study examines the perceived trust of different COVID-19 information sources (health professionals, academic institutions, government agencies, news media, social media, family, and friends) and sharing of COVID-19 information in China. Specifically, it investigates how beliefs about sharing and emotions mediate the effects of perceived source trust on source-specific information sharing intentions. Results suggest that health professionals, academic institutions, and government agencies are trusted sources of information and that people share information from these sources because they think doing so will increase disease awareness and promote disease prevention. People may also choose to share COVID-19 information from news media, social media, and family as they cope with anxiety, anger, and fear. Taken together, a better understanding of the distinct psychological mechanisms underlying health information sharing from different sources can help contribute to more effective sharing of information about COVID-19 prevention and to manage negative emotion contagion during the pandemic.

The attention mechanism has been widely applied to enhance pedestrian representation for person re-identification in videos. However, most existing methods learn the spatial and temporal attention separately, and thus ignore the correlation between them. In this work, we propose a novel Adaptive Spatio-Temporal Attention Network (ASTA-Net) to adaptively aggregate the spatial and temporal attention features into discriminative pedestrian representation for person re-identification in videos. Specifically, multiple Adaptive Spatio-Temporal Fusion modules within ASTA-Net are designed for exploring precise spatio-temporal attention on multi-level feature maps. They first obtain the preliminary spatial and temporal attention features via the spatial semantic relations for each frame and temporal dependencies among inconsecutive frames, then adaptively aggregate the preliminary attention features on the basis of their correlation. Moreover, an Adjacent-Frame Motion module is designed to explicitly extract motion patterns according to the feature-level variation among adjacent frames. Extensive experiments on the three widely-used datasets, i.e., MARS, iLIDS-VID and PRID2011, have demonstrated the effectiveness of the proposed approach.

Person search has recently gained increasing attention as the novel task of localizing and identifying a target pedestrian from a gallery of non-cropped scene images. Its performance depends on accurate person detection and re-identification simultaneously by learning effective representations. In this work, we propose a novel dual context-aware refinement network (DCRNet) for person search, which jointly explores two kinds of contexts including intra-instance context and inter-instance context to learn discriminative representation. Specifically, an intra-instance context module is designed to refine the representation for the bounding box of a pedestrian by leveraging its surrounding regions covering the same pedestrian and its accessories, which contain abundant complementary visual appearance of pedestrians. Moreover, an inter-instance context module is proposed to expand the instance-level feature for the bounding box of a pedestrian, by utilizing the rich scene contexts of neighboring co-travelers across images. These two modules are built on top of a joint detection and feature learning framework, i.e., Faster R-CNN. Extensive experimental results on two challenging datasets have demonstrated the effectiveness of DCRNet with significant performance improvements over state-of-the-art methods.