Search

Support vector machines (SVM) have drawn wide attention for the last two decades due to its extensive applications, so a vast body of work has developed optimization algorithms to solve SVM with various soft-margin losses. To distinguish all, in this paper, we aim at solving an ideal soft-margin loss SVM: L0/1 soft-margin loss SVM (dubbed as L0/1-SVM). Many of the existing (non)convex soft-margin losses can be viewed as one of the surrogates of the L0/1 soft-margin loss. Despite its discrete nature, we manage to establish the optimality theory for the L0/1-SVM including the existence of the optimal solutions, the relationship between them and P-stationary points. These not only enable us to deliver a rigorous definition of L0/1 support vectors but also allow us to define a working set. Integrating such a working set, a fast alternating direction method of multipliers is then proposed with its limit point being a locally optimal solution to the L0/1-SVM. Finally, numerical experiments demonstrate that our proposed method outperforms some leading classification solvers from SVM communities, in terms of faster computational speed and a fewer number of support vectors. The bigger the data size is, the more evident its advantage appears.

Icing significantly affects the performance of wind turbines in terms of power loss and structural degradation, and an effective blade icing diagnosis is the prerequisite to achieve the optimal control of wind turbines to mitigate such icing influence. However, current icing diagnostic methods lack consideration of fundamental icing physics and have limited generalizability to large-scale applications. To address such challenges, in the present study, we aim to propose an effective and robust blade icing diagnostic method for wind turbines. Specifically, hybrid features that fully consider both short-term and long-term icing influence are extracted based on the underlying icing physics. Such features are used to build a Stacked-XGBoost model (i.e., based on a combination of stacking ensemble learning algorithm and XGBoost machine learning algorithm) to achieve blade icing diagnosis. The proposed method is evaluated at two wind farms and further compared with three single algorithm-based models (i.e., random forest, support vector machine and XGBoost algorithms). The results show that the hybrid features significantly enhance the similarity between different datasets and the Stacked-XGBoost algorithm achieves a higher diagnostic accuracy and a better generalizability compared to the single-algorithm-based models.

Semantic segmentation of remotely sensed imagery plays a critical role in many real-world applications, such as environmental change monitoring, precision agriculture, environmental protection, and economic assessment. Following rapid developments in sensor technologies, vast numbers of fine-resolution satellite and airborne remote sensing images are now available, for which semantic segmentation is potentially a valuable method. However, because of the rich complexity and heterogeneity of information provided with an ever-increasing spatial resolution, state-of-the-art deep learning algorithms commonly adopt complex network structures for segmentation, which often result in significant computational demand. Particularly, the frequently-used fully convolutional network (FCN) relies heavily on fine-grained spatial detail (fine spatial resolution) and contextual information (large receptive fields), both imposing high computational costs. This impedes the practical utility of FCN for real-world applications, especially those requiring real-time data processing. In this paper, we propose a novel Attentive Bilateral Contextual Network (ABCNet), a lightweight convolutional neural network (CNN) with a spatial path and a contextual path. Extensive experiments, including a comprehensive ablation study, demonstrate that ABCNet has strong discrimination capability with competitive accuracy compared with state-of-the-art benchmark methods while achieving significantly increased computational efficiency. Specifically, the proposed ABCNet achieves a 91.3% overall accuracy (OA) on the Potsdam test dataset and outperforms all lightweight benchmark methods significantly. The code is freely available at https://github.com/lironui/ABCNet .

Fatigue damage assessment is critical for the structural design of the wind turbine nacelle chassis. However, existing fatigue damage indicators and assessment method cannot be both fast and accurate. In this paper, we propose a quantitative support vector regression-based fatigue damage assessment (SVR-FDA) method. First, the equivalent stress amplitude (ESA) is defined to simplify the fatigue damage indicator. Second, we establish the pre-calculated ESA database, including the ESAs of the wind turbine nacelle chassis under many varying wind flow conditions. Finally, based on the pre-calculated ESA database, we establish the SVR-FDA model, which can calculate the ESA of any given wind flow condition. A wind turbine nacelle chassis fatigue damage dataset, released by Goldwind, was applied to validate the proposed method. The results demonstrated that the SVR-FDA yielded the highest assessment accuracy for the lifetime ESA, as compared with four popular machine learning algorithms, including the least absolute shrinkage and selection operator, random forest, extreme gradient boosting, and deep neural network.

To solve problems in 3D modelling, including unclear 3D edges in point clouds, lack of geometric semantics, confusing topological relations between 3D models, and low degree of automation in traditional model-driven 3D modelling, a method for high-accuracy and automatic monomer 3D modelling must be developed. This paper is the first to propose a 3D modelling strategy, called combining data-and-model-driven 3D modelling (CDMD3DM), for small regular objects in indoor scenes using RGB-D data. The proposed method’s workflow is as follows: generation of initial 3D point cloud data using data-driven Kinect v2; segmentation of point cloud data, based on deep learning, that improves the accuracy and automation of geometric model recognition; definition of the initial model-driven parameters based on the instance segmentation results; optimization of geometric model parameters, based on generalized point photogrammetry theory, to generate monomer models in indoor scenes to overcome the shortcomings of confusing topological relationships and inaccurate 3D model edges; and finally, fusion of the results of data-driven and model-driven 3D modelling. The experimental results demonstrate that CDMD3DM is feasible, automatic, more accurate, more reliable, semantically richer and capable of producing clearer topological relationships than current 3D modelling results using indoor RGB-D data. These outcomes promote interdisciplinary integration between computer vision and photogrammetry.

COVID-19 has undergone several mutations and is still spreading in most countries now. PA has positive benefits in the prevention of COVID-19 infection and counteracting the negative physical and mental effects caused by COVID-19. However, relevant evidence has indicated a high prevalence of physical inactivity among the general population, which has worsened due to the outbreak of the pandemic, and there is a severe lack of exercise guidance and mitigation strategies to advance the knowledge and role of PA to improve physical and mental health in most countries during the epidemic. This study surveyed the effects of COVID-19 on PA in Chinese residents during the pandemic and provided important reference and evidence to inform policymakers and formulate policies and planning for health promotion and strengthening residents' PA during periods of public health emergencies. ANOVA, Kolmogorov-Smirnov, the chi-square test and Spearman correlation analysis were used for statistical analysis. A total of 14,715 participants were included. The results show that nearly 70% of Chinese residents had inadequate PA (95%CI 58.0%-82.19%) during the COVID-19 outbreak, which was more than double the global level (27.5%, 95%CI 25.0%-32.2%). The content, intensity, duration, and frequency of PA were all affected during the period of home isolation, and the types of PA may vary among different ages. The lack of physical facilities and cultural environment is the main factor affecting PA. However, there was no significant correlation between insufficient PA and the infection rate. During the period of home isolation and social distance of epidemic prevention, it is necessary to strengthen the scientific remote network monitoring and guidance for the process of PA in China.

The highly dynamic nature of agro-ecosystems in space and time usually leads to high intra-class variance and low inter-class separability in the fine spatial resolution (FSR) remotely sensed imagery. This makes traditional classifiers essentially relying on spectral information for crop mapping from FSR imagery an extremely challenging task. To mine effectively the rich spectral and spatial information in FSR imagery, this paper proposed a Scale Sequence Object-based Convolutional Neural Network (SS-OCNN) that classifies images at the object level by taking segmented objects (crop parcels) as basic units of analysis, thus, ensuring that the boundaries between crop parcels are delineated precisely. These segmented objects were subsequently classified using a CNN model integrated with an automatically generated scale sequence of input patch sizes. This scale sequence can fuse effectively the features learned at different scales by transforming progressively the information extracted at small scales to larger scales. The effectiveness of the SS-OCNN was investigated using two heterogeneous agricultural areas with FSR SAR and optical imagery, respectively. Experimental results revealed that the SS-OCNN consistently achieved the most accurate classification results. The SS-OCNN, thus, provides a new paradigm for crop classification over heterogeneous areas using FSR imagery, and has a wide application prospect.

Background Fibrinogen (FIB) is an independent risk factor for mortality and cardiovascular events in the general population. However, the relationship between FIB and long-term mortality among CAD patients undergoing PCI remains unclear, especially in individuals complicated with diabetes mellitus (DM) or prediabetes (Pre-DM). Methods 6,140 patients with CAD undergoing PCI were included in the study and subsequently divided into three groups according to FIB levels (FIB-L, FIB-M, FIB-H). These patients were further grouped by glycemic status [normoglycemia (NG), Pre-DM, DM]. The primary endpoint was all-cause mortality. The secondary endpoint was cardiac mortality. Results FIB was positively associated with hemoglobin A1c (HbA1c) and fasting blood glucose (FBG) in CAD patients with and without DM (P Conclusions FIB was independently associated with long-term all-cause and cardiac mortality among CAD patients undergoing PCI, especially in those with DM and Pre-DM. FIB test may help to identify high-risk individuals in this specific population.

Objective: Cognitive dysfunction often accompanies diabetes. Both hypoglycemia and hyperglycemia cause cognitive dysfunctions. However, the underlying pathophysiology remains unclear. Recent evidence show that ferroptosis primarily triggers nerve cell death, Alzheimer's disease (AD), Huntington’s disease (HD), and Parkinson's disease (PD). We purposed to investigate whether ferroptosis is a vital pathogenic pathway in diabetes-induced cognitive dysfunction. Methods and results: Type 1 diabetic rat model was created by intraperitoneal injection of streptozotocin (STZ). Significant cognitive dysfunction was observed in the diabetic rats as evidenced by increase in latency period to find a hidden platform and decreased cumulative time spent in the target quadrant in the Morris water maze test. We detected the amplitude of low-frequency fluctuation (ALFF) of the BOLD (Blood Oxygenation Level-Dependent) signal using resting-state functional magnetic resonance imaging (rs-fMRI). Consequently, we found that the ALFF values, as well as the T2 relaxation time of the bilateral hippocampus, were reduced in Type 1 diabetic rats. We detected Fe2+ level and lipid peroxidation products (MDA and 4-HNE) in the hippocampus. Mitochondria and neuron injury in the STZ-induced diabetic rats were determined using a Transmission Electron Microscope and Nissl body staining. Iron overload and ferroptosis were detected in the hippocampus. Furthermore, mRNA microarray analysis revealed 201 dysregulated mRNAs in STZ-induced type 1 diabetes (T1D). Pathway enrichment analyses indicated that differentially expressed mRNAs associated-coding genes were associated with ferroptosis. Among ferroptosis signaling pathway genes, Slc40a1 gene (ferroportin) was downregulated.Conclusion: We show that ferroptosis is associated with diabetic cognitive dysfunction and Slc40a1 mediates ferroptosis in Type 1 diabetes.

Model Agnostic Meta Learning (MAML) has become the most representative meta learning algorithm to solve few-shot learning problems. This paper mainly discusses MAML framework, focusing on the key problem of solving few-shot learning through meta learning. However, MAML is sensitive to the base model for the inner loop, and training instability occur during the training process, resulting in an increase of the training difficulty of the model in the process of training and verification process, causing degradation of model performance. In order to solve these problems, we propose a multi-stage loss optimization meta-learning algorithm. By discussing a learning mechanism for inner and outer loops, it improves the training stability and accelerates the convergence for the model. The generalization ability of MAML has been enhanced.

The increasing popularity of Apache Spark has attracted many users to put their data into its ecosystem. On the other hand, it has been witnessed in the literature that Spark is slow when it comes to distributed machine learning (ML). One resort is to switch to specialized systems such as parameter servers, which are claimed to have better performance. Nonetheless, users have to undergo the painful procedure of moving data into and out of Spark. In this paper, we investigate performance bottlenecks of MLlib (an official Spark package for ML) in detail, by focusing on analyzing its implementation of stochastic gradient descent (SGD)—the workhorse under the training of many ML models. We show that the performance inferiority of Spark is caused by implementation issues rather than fundamental flaws of the bulk synchronous parallel (BSP) model that governs Spark’s execution: we can significantly improve Spark’s performance by leveraging the well-known “model averaging” (MA) technique in distributed ML. Indeed, model averaging is not limited to SGD, and we further showcase an application of MA to training latent Dirichlet allocation (LDA) models within Spark. Our implementation is not intrusive and requires light development effort. Experimental evaluation results reveal that the MA-based versions of SGD and LDA can be orders of magnitude faster compared to their counterparts without using MA.

Whether routine assessment of FT3/FT4 ratio in euthyroid patients with three-vessel disease (3VD) could help identify high-risk individuals remains unclear. This study evaluated the relationship between FT3/FT4 ratio and long-term clinical outcomes in this specific population.This study included 2106 euthyroid patients with 3VD (stenoses of ≥50% in right coronary artery, left circumflex and left anterior descending). Patients were categorized into three groups according to tertiles of FT3/FT4 ratio (Q12.58,n = 704; 2.2 ≤ Q22.58, n = 706; Q32.22, n = 696). The median follow-up time was 5.3 years, during which 206 deaths and 332 MACCEs (consisting of all-cause death, myocardial infarction, and stroke) occurred. Compared with the other two groups, patients with low level of FT3/FT4 ratio tended to be female, older, diabetic, and had significantly higher incidences of all-cause death, cardiac death and MACCE (all P 0.05). Cox regression analysis showed that patients with low level of FT3/FT4 ratio had higher risks of long-term cardiac death (adjusted HR = 1.87, 95% CI 1.06-3.28, P = 0.030) and MACCE (adjusted HR = 1.43, 95% CI 1.07-1.93, P = 0.017) than those with high level of FT3/FT4 ratio. Subgroup analysis showed there was a significant interaction between FT3/FT4 ratio and age (≥65 years vs.65 years) for MACCE (P = 0.029).Low level of FT3/FT4 ratio is independently associated with an increased risk of long-term cardiac death and MACCE in euthyroid patients with 3VD. Routine assessment of FT3/FT4 ratio might be helpful to identify high-risk individuals in this specific population.

The autologous conchal cartilage of good elasticity is easy to harvest, thus is often used in nasal tip plasty of East Asians. However, the operation techniques vary a lot among different surgeons. This article aims to introduce 3 techniques commonly used in clinical practice.One hundred three patients were included in this study and divided into 3 groups according to the shape of the nasal tip during 2017 to 2019. The patients were followed up for at least 6 months. All patients were measured with Standardized Cosmesis and Health Nasal Outcomes Survey (SCHNOS). ImageJ software was used to measure and calculate the projection ratio for lateral position standard image, and columella-labial angle (CLA) was measured.In group 1, SCHNOS for nasal obstruction (SCHNOS-O) score were 13.23 ± 7.61 and 14.49 ± 10.55 (P0.05); SCHNOS for nasal cosmesis (SCHNOS-C) score were 66.55 ± 31.23 and 21.73 ± 18.91 (P0.001); projection ratio were 0.51 ± 0.04 and 0.57 ± 0.05 (P0.001); CLA were 91.02° ± 5.67° and 93.55° ± 4.64° (P0.05), preoperative and postoperative, respectively. In group 2, SCHNOS-O score were 16.76 ± 13.44 and 15.44 ± 9.26 (P0.05); SCHNOS-C score were 61.87 ± 44.14 and 19.16 ± 29.37 (P0.001); projection ratio were 0.50 ± 0.05 and 0.58 ± 0.03 (P0.001); CLA were 92.32° ± 6.39° and 94.86° ± 5.96°(P0.05), preoperative and postoperative, respectively. In group 3, SCHNOS-O score were 15.25 ± 7.83 and 17.73 ± 11.66 (P0.05); SCHNOS-C score were 52.03 ± 33.38 and 17.73 ± 11.66 (P0.001); projection ratio were 0.53 ± 0.05 and 0.57 ± 0.02 (P0.001); CLA were 91.65° ± 5.75° and 93.21° ± 5.38° (P0.05), preoperational and postoperational, respectively.None of these 3 techniques cause or aggravate nasal obstruction, and all of them can achieve high cosmetic satisfaction. Technique 1 and technique 2 are suitable for the patients with moderate and moderate to severe short nose that is common in East Asia, which can better increase the protrusion of the tip and length of the nose. Technique 3 is suitable for the patients with better shape of the nose, who need to slightly increase the protrusion of the nasal tip and increase the upward rotation.

Background Moderate/severe coronary artery calcification (CAC) predicts worse clinical outcomes in patients undergoing percutaneous coronary intervention (PCI). However, to date most studies have been modest in size and with limited follow-up. We aimed to assess the association between calcification severity and long-term clinical outcomes in a large cohort undergoing PCI.Methods and Results:In total, 10,068 consecutive patients who underwent PCI at Fuwai Hospital were enrolled in this prospective observational study. Patients were categorized as none/mild or moderate/severe CAC according to the severity of the target lesion by visual assessment of coronary angiography. Major adverse cardiovascular events (MACE), a composite event of death, myocardial infarction and revascularization, at 5 years were assessed. None/mild CAC was observed in 8,229 (81.7%) patients, and moderate/severe CAC was observed in 1,839 (18.3%) patients. Patients with moderate/severe CAC had a significantly higher rate of 5-year unplanned revascularization (15.2% vs. 13.2%, P=0.022) and MACE (20.7% vs. 17.9%, P=0.005). After propensity score matching, the moderate/severe CAC group still had a higher rate of 5-year unplanned revascularization (15.2% vs. 12.6%, P=0.019). Cox regression analysis using clinically significant variables revealed moderate/severe calcification was independently associated with higher risk of 2-year unplanned target vessel revascularization (hazard ratio (HR)=1.287, 95% confidence interval (CI): 1.036-1.600, P=0.023) and MACE (HR=1.242, 95% CI: 1.039-1.484, P=0.017), but not 5-year unplanned revascularization and MACE. Conclusions In patients undergoing PCI, moderate/severe coronary calcification increases the risk of long-term MACE.

ObjectiveElevated levels of N-terminal pro-brain natriuretic peptide (NT-proBNP) are associated with heart failure-related death in hypertrophic cardiomyopathy (HCM), but the relationship between NT-proBNP level and sudden cardiac death (SCD) in HCM remains undefined.MethodsThe study prospectively enrolled 977 unrelated patients with HCM with available NT-proBNP results who were prospectively enrolled and followed for 3.0±2.1 years. The Harrell’s C-statistic under the receiver operating characteristic curve was calculated to evaluate discrimination performance. A combination model was constructed by adding NT-proBNP tertiles to the HCM Risk-SCD model. The correlation between log NT-proBNP level and cardiac fibrosis as measured by late gadolinium enhancement (LGE) or Masson’s staining was analysed.ResultsDuring follow-up, 29 patients had SCD. Increased log NT-proBNP levels were associated with an increased risk of SCD events (adjusted HR 22.27, 95% CI 10.93 to 65.63, pConclusionNT-proBNP is an independent predictor of SCD in patients with HCM and may help with risk stratification of this disease.

Background Scalp combing, as an ancient method of health care, has been used for thousands of years in traditional Chinese medicine. Although this method is considered to be beneficial for the blood circulation of the head, the underlying mechanisms remain unclear. Methods Both human participants and mice were used in this study. In participants, the scalp was stimulated by combing continuously for 5 min, and the temperature was measured using infrared thermal imaging before and after stimulation. In mice, the temperature was determined before and at 5, 15, and 30 min after a 5-min scalp mechanical stimulation (SMS). Additionally, the vasculature of the mice was labeled with retro-orbital fluorescein isothiocyanate-dextran injection, and the capillaries were observed directly under a confocal microscope. Using in vivo CLARITY imaging and the spectrofluorometric detection of Evans Blue dye extravasation, the blood–brain barrier permeability was assessed. Results SMS increased the temperature of the left ear significantly in human (P = .0247) while can slightly increase the temperature of the right ear and the face without significant difference (P > .05). Moreover, SMS can significantly slow the decrease in the temperature of the external auditory canal at 5 min (P = .0153) and in body temperature at 15 min (P = .0295) after SMS whereas no significant difference in body temperature at 30 min (P > .05) compared with control mice. Furthermore, capillaries of the ear with a diameter of less than 8 μm were significantly dilated (P = .0006) following SMS and the number of dextran dots was higher at 15 min (P > .05) and 30 min (F = 10.98, P = .037) after SMS intervention compared with control mice. Conclusion Our study provides strong evidence to support the notion that scalp combing can improve extra- and intracranial blood circulation under healthy conditions.

AIM: To compare the safety and efficacy of the intravitreal injection of conbercept (IVC) and triamcinolone acetonide (IVTA) for macular edema (ME) secondary to branch retinal vein occlusion (BRVO). METHODS: A prospective, randomized clinical study. Patients with ME secondary to BRVO were randomly assigned to either IVC group or IVTA group at a ratio of 2:1 and a 12-month follow-up was performed. The efficacy outcome measures included the mean changes and differences in best corrected visual acuity (BCVA) and the central retinal thickness (CRT). The safety profiles and the mean retreatment intervals were also compared. RESULTS: There was no statistically significant difference of baseline between the two groups (IVC group, n=36; IVTA group, n=17). At 12mo, the BCVA letters improved by 27.31±18.36 in the IVC group, and 13.53±11.37 in the IVTA group (P=0.0004). CRT reduction was 253.33±163.69 and 150.24±134.32 µm, respectively (P=0.0034). The mean BCVA in the IVC group was superior to that of the IVTA group for months 6-12 (P

Gefitinib (Iressa), is a selective epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor (TKI), used in the targeted treatment of locally advanced or metastatic non-small cell lung cancer (NSCLC). Skin toxicity is the major adverse effect observed in patients treated with EGFR-targeted TKIs such as gefitinib and erlotinib. To date, a corresponding skin animal model has not been established to address the mechanisms of these effects. Therefore, we analyzed the skin rash phenotype and its pathological features in Brown Norway (BN) rats treated with gefitinib 2.5 mg, 5.0 mg, or 10 mg/100 g/day for 4 weeks. We found that treatment with gefitinib led to weight loss, rash, itching, and hair loss in a dose-dependent manner. We also investigated the skin pathology and found that the animal model showed thickening of the epidermis, loss of moisture, and apoptosis of keratinocytes. Immunohistochemistry, flow cytometry, and analysis of monocytes and leukocytes in the blood revealed increased macrophage infiltration was associated with the cutaneous toxicities induced by gefitinib in the BN rats. Finally, we found that gefitinib-induced cutaneous toxicity is significantly associated with three inflammatory cytokines known to be secreted by activated macrophages, TREM-1, CINC-2, and CINC-3.

Pan-sharpening aims to obtain high-resolution multi-spectral images by fusing panchromatic images and low-resolution multi-spectral images though reasonable rules. This paper proposed a novel generative adversarial network for pan-sharpening, which utilizes the supplemented spectral information from low-resolution multi-spectral images and the enhanced structural information from panchromatic images to generate high-resolution multi-spectral images. Firstly, the forward differential operator is used to extract the spatial structural information of the panchromatic image both in the horizontal and vertical directions. Secondly, an architecture of generative adversarial network is designed. The enhanced structural information generated by the accumulation of the structural information of the two directions is added to the image fusion process in generator and the discriminating process in discriminator, and a new optimization objective is designed accordingly. What is more, the low-resolution multi-spectral image is added to the convolution process in the generator as a supplement to the spectral information. Finally, in order to obtain better image generation effect, a special objective function of the generator is designed, which adds a unique relationship to reduce the loss of spatial structural information and spectral information of fused images. Experiments on QuickBird and WorldView-3 satellites datasets show that the proposed method can generate high quality fused images and is better than most advanced methods in both objective indicators and intuitive observations.

Remote sensing scene classification plays a critical role in a wide range of real-world applications. Technically, however, scene classification is an extremely challenging task due to the huge complexity in remotely sensed scenes, and the difficulty in acquiring labelled data for model training such as supervised deep learning. To tackle these issues, a novel semi-supervised ensemble framework is proposed here using the self-training hierarchical prototype-based classifier as the base learner for chunk-by-chunk prediction. The framework has the ability to build a powerful ensemble model from both labelled and unlabelled images with minimum supervision. Different feature descriptors are employed in the proposed ensemble framework to offer multiple independent views of images. Thus, the diversity of base learners is guaranteed for ensemble classification. To further increase the overall accuracy, a novel cross-checking strategy was introduced to enable the base learners to exchange pseudo-labelling information during the self-training process, and maximize the correctness of pseudo-labels assigned to unlabelled images. Extensive numerical experiments on popular benchmark remote sensing scenes demonstrated the effectiveness of the proposed ensemble framework, especially where the number of labelled images available is limited. For example, the classification accuracy achieved on the OPTIMAL-31, PatternNet and RSI-CB256 datasets was up to 99.91%, 98. 67% and 99.07% with only 40% of the image sets used as labelled training images, surpassing or at least on par with mainstream benchmark approaches trained with double the number of labelled images.

Semantic segmentation of remotely sensed images plays an important role in land resource management, yield estimation, and economic assessment. U-Net, a deep encoder-decoder architecture, has been used frequently for image segmentation with high accuracy. In this letter, we incorporate multiscale features generated by different layers of U-Net and design a multiscale skip connected and asymmetric-convolution-based U-Net (MACU-Net), for segmentation using fine-resolution remotely sensed images. Our design has the following advantages: (1) the multiscale skip connections combine and realign semantic features contained in both low-level and high-level feature maps; (2) the asymmetric convolution block strengthens the feature representation and feature extraction capability of a standard convolution layer. Experiments conducted on two remotely sensed data sets captured by different satellite sensors demonstrate that the proposed MACU-Net transcends the U-Net, U-Netpyramid pooling layers (PPL), U-Net 3+, among other benchmark approaches. Code is available at https://github.com/lironui/MACU-Net.

For the object-based classification of high resolution remote sensing images, many people expect that introducing deep learning methods can improve then classification accuracy. Unfortunately, the input shape for deep neural networks (DNNs) is usually rectangular, whereas the shapes of the segments output by segmentation methods are usually according to the corresponding ground objects; this inconsistency can lead to confusion among different types of heterogeneous content when a DNN processes a segment. Currently, most object-based methods utilizing convolutional neural networks (CNNs) adopt additional models to overcome the detrimental influence of such heterogeneous content; however, these heterogeneity suppression mechanisms introduce additional complexity into the whole classification process, and these methods are usually unstable and difficult to use in real applications. To address the above problems, this paper proposes a simplified object-based deep neural network (SO-DNN) for very high resolution remote sensing image classification. In SO-DNN, a new segment category label inference method is introduced, in which a deep semantic segmentation neural network (DSSNN) is used as the classification model instead of a traditional CNN. Since the DSSNN can obtain a category label for each pixel in the input image patch, different types of content are not mixed together; therefore, SO-DNN does not require an additional heterogeneity suppression mechanism. Moreover, SO-DNN includes a sample information optimization method that allows the DSSNN model to be trained using only pixel-based training samples. Because only a single model is used and only a pixel-based training set is needed, the whole classification process of SO-DNN is relatively simple and direct. In experiments, we use very high-resolution aerial images from Vaihingen and Potsdam from the ISPRS WG II/4 dataset as test data and compare SO-DNN with 6 traditional methods: O-MLP, O+CNN, OHSF-CNN, 2-CNN, JDL and U-Net. Compared with the best-performing method among these traditional methods, the classification accuracy of SO-DNN is improved by up to 7.71% and 10.78% for single images from Vaihingen and Potsdam, respectively, and the average classification accuracy is improved by 2.46% and 2.91% for the Vaihingen and Potsdam images, respectively. SO-DNN relies on fewer models and easier-to-obtain samples than traditional methods, and its stable performance makes SO-DNN more valuable for practical applications.

Low dopamine levels may cause depressive symptoms. Dopamine is also involved in sexual behavior. Rotigotine is a nonergolinic dopamine agonist. Fluoxetine, an antidepressant that acts as a selective serotonin (5-HT) reuptake inhibitor, may cause moderate or severe sexual dysfunction. This study aims to investigate the effects of rotigotine-loaded microspheres (RoMS) and rotigotine on fluoxetine-induced impairment of sexual function and their efficacy in depression-model rats. Rats with depressive-like behavior, induced by bilateral olfactory bulbectomy, were treated intragastrically with fluoxetine and co-administered RoMS or rotigotine subcutaneously. Then, copulatory behavior and open field tests were conducted. Serum luteinizing hormone and testosterone levels were assayed with enzyme-linked immunosorbent assay kits. The concentrations of 5-HT, dopamine, and norepinephrine were measured in the raphe nucleus and amygdala. The results showed that sexual function was decreased in olfactory bulbectomy rats and significantly deteriorated by fluoxetine. Co-administration of RoMS partly reversed the fluoxetine-induced impairment of sexual function, but rotigotine administration did not produce any improvement. Hyperactivity in olfactory bulbectomy rats was significantly attenuated by fluoxetine but was not influenced by co-administration of RoMS. Compared with the fluoxetine group, RoMS increased the testosterone, luteinizing hormone, dopamine, and norepinephrine levels. These findings indicated that RoMS improved the fluoxetine-induced impairment of sexual function and did not affect its antidepressant efficacy in depressive rats, which provides a potential treatment for patients with depression that can reduce the possibility of sexual dysfunction. Additionally, co-administration of fluoxetine with RoMS may be beneficial for Parkinson's disease patients with depression.

The agricultural landscape can be interpreted at different semantic levels, such as fine low-level crop (LLC) classes (e.g., Wheat, Almond, and Alfalfa) and broad high-level crop (HLC) classes (e.g., Winter crops, Tree crops, and Forage). The LLC and HLC are hierarchically correlated with each other, but such intrinsically hierarchical relationships have been overlooked in previous crop classification studies in remote sensing. In this research, a novel Iterative Deep Learning (IDL) framework was proposed for the classification of complex agricultural landscapes using remotely sensed imagery. The IDL adopts an object-based convolutional neural network (OCNN) as the basic classifier for both the LLC and HLC classifications, which has the advantage of maintaining precise crop parcel boundaries. In IDL, the HLC classification implemented by the OCNN is conditional upon the LLC classification probabilities, whereas the HLC probabilities combined with the original imagery are, in turn, re-used as inputs to the OCNN to enhance the LLC classification. Such an iterative updating procedure forms a Markov process, where both the LLC and HLC classifications are refined and evolve collaboratively. The effectiveness of the IDL was tested on two heterogeneous agricultural fields using fine spatial resolution (FSR) SAR and optical imagery. The experimental results demonstrate that the iterative process of IDL helps to resolve contradictions within the class hierarchies. The new proposed IDL consistently increased the accuracies of both the LLC and HLC classifications with iteration, and achieved the highest accuracies for each at four iterations. The average overall accuracies were 88.4% for LLC and 91.2% for HLC, for both study sites, far greater than the accuracies of the state-of-the-art benchmarks, including the pixel-wise CNN (81.7% and 85.9%), object-based image analysis (OBIA) (84.0% and 85.8%), and OCNN (84.0% and 88.4%). To the best of our knowledge, the proposed model is the first to identify and use the relationship between the class levels in an ontological hierarchy in a remote sensing classification process. It is applied here to increase progressively the accuracy of classification at two levels for a complex agricultural landscape. As such IDL represents an entirely new paradigm for remote sensing image classification. Moreover, the promising results demonstrate the great potential of the proposed IDL with wide application prospect.

Machine Learning methods have been adopted for a wide range of real-world applications, ranging from social networks, online image/video-sharing platforms, and e-commerce to education, healthcare, etc. However, in practice, a large amount of effort is required to tune several components of machine learning methods, including data representation, hyperparameter, and model architecture, in order to achieve a good performance. To alleviate the required tunning efforts, Automated Machine Learning (AutoML), which can automate the process of applying machine learning methods, has been studied in both academy and industry recently. In this tutorial, we will introduce the main research topics of AutoML, including Hyperparameter Optimization, Neural Architecture Search, and Meta-Learning. Two emerging topics of AutoML, Automatic Feature Generation and Machine Learning Guided Database, will also be discussed since they are important components for real-world applications. For each topic, we will motivate it with application examples from industry, illustrate the state-of-the-art methodologies, and discuss some future research directions based on our experience from industry and the trends in academy.

Amorphous materials have low mobility due to their nature of disorder. Surprisingly, some disordered materials showed photocurrent amplification not by conventional photoconductive gain. Recently, amorphous Silicon (a-Si) photodiodes with thin a-Si layer (∼40 nm) have shown a gain-bandwidth product of over 2 THz with very low excess noise and also have been used as a gain media in a cascaded system with single photon sensitivity. To unveil the true gain mechanism, we performed theoretical modeling and numerical analysis along with experimental data at different frequencies. We show evidence of highly effective carrier multiplication process within a-Si as the primary gain mechanism, especially at high frequency. We also show presence of trap-induced junction modulation at much lower frequency. We modeled the gain mechanism in a-Si by solving the transport equations including dynamics of defect states and carrier multiplication via the local field model. We further justified the application of local field model for thin a-Si, based on the property that in a-Si, the mean-free path for energy relaxation is orders of magnitude greater than the mean-free-path for momentum relaxation. The analysis further suggests that the carrier multiplication process in thin a-Si can be much more efficient than in thick a-Si, even stronger than single crystalline Si in some cases. Although seemingly counter intuitive, this is consistent with the proposed cycling excitation process where the localized states in the bandtails of disordered materials such as a-Si relax the k -selection rule and increase the rate of carrier multiplication.

The genetic factors related to early-onset hypertension are largely unknown. This study aimed to determine the spectrum of steroid metabolism gene variants and the clinical relationships of these variants to phenotypes in Chinese patients with early-onset hypertension. A total of 306 consecutive early-onset hypertensive patients were recruited. All coding exons and flanking intronic regions of KCNJ5, CYP11B1, and CYP17A1 were sequenced. Long-distance polymerase chain reaction was used to search for a CYP11B1/CYP11B2 chimeric gene. Pedigree investigations and genotype-phenotype analyses were performed for patients with rare variants. Nine rare variants were detected in eight patients (2.6%), but no CYP11B1/CYP11B2 chimeric gene was identified. One patient and two of her siblings were found to carry compound heterozygous mutations (C183Y and T390R) in CYP17A1 and were eventually diagnosed with atypical congenital adrenal hyperplasia. Patients with rare variants had younger ages of onset [17 (16, 20) vs. 30 (23, 35) years old, p = 0.010] and higher systolic blood pressure (148.5 ± 9.6 vs. 137.9 ± 17.8 mmHg, p = 0.021) than those without rare variants. Additionally, the patients and their relatives carrying rare variants exhibited increased serum free corticosterone [230.4 (7.4, 533.0) vs. 1.9 (0.9, 6.7)ng/ml, p = 0.001] and 11-deoxycorticosterone [16.16 (0.59, 33.23) vs. 0.77 (0.41, 0.96)ng/ml, p = 0.038] levels. Genetic testing is useful for the etiologic diagnosis of early-onset hypertension. Rare variants in steroid metabolism genes were associated with more severe clinical expression and abnormal circulating steroid metabolites in patients with early-onset hypertension.

In wireless sensor networks (WSNs), sensor readings have spatiotemporal correlation, and wireless links are unreliable. To balance the energy-performance trade-off and reduce the impact of packet loss, a novel approach is proposed that combines Kronecker compressed sensing (KCS) and cluster topology to exploit spatial and temporal correlations simultaneously. The head nodes generate sparse sub-measurement matrices based on gathered data to avoid measuring nodes that cannot successfully transmit readings. The sink constructs those matrices as a block diagonal matrix (BDM) to utilize the spatial correlation among clusters. Numerical results show that this scheme effectively balances the energy-performance trade-off and maintains a high reconstruction accuracy with packet loss.

Database engines are starting to incorporate machine learning (ML) functionality as part of their repertoire. Machine learning algorithms, however, have very different characteristics than those of relational operators. In this demonstration, we explore the challenges that arise when integrating generalized linear models into a database engine and how to incorporate hardware accelerators into the execution, a tool now widely used for ML workloads. The demo explores two complementary alternatives: (1) how to train models directly on compressed/encrypted column-stores using a specialized coordinate descent engine, and (2) how to use a bitwise weaving index for stochastic gradient descent on low precision input data. We present these techniques as implemented in our prototype database doppioDB 2.0 and show how the new functionality can be used from SQL.

Aims Risk assessment and treatment stratification for three-vessel disease (3VD) remain challenging. N-terminal pro-B-type natriuretic peptide (NT-proBNP) is an established biomarker for prognostication and treatment in heart failure. The present study aimed to evaluate the prognostic value of NT-proBNP beyond the SYNTAX score II (SSII), and its association with long-term outcome after three strategies [percutaneous coronary intervention (PCI), coronary artery bypass grafting (CABG), and medical therapy (MT)] in a cohort of patients with 3VD. Methods and results A total of 6597 patients with available baseline NT-proBNP data were included in the study. Baseline, procedural, and follow-up data were collected. The primary endpoint was all-cause death. Secondary endpoints included cardiac death and major adverse cardiac and cerebrovascular events (MACCE), a composite of death, myocardial infarction, and stroke. During a median follow-up of 7.0 years, higher NT-proBNP levels were strongly associated with increased risks of all-cause death, cardiac death, and MACCE (all adjusted P Conclusion N-terminal pro-BNP is a potential biomarker for risk stratification and therapeutic decision-making in patients with 3VD. Further randomized studies are needed to confirm these findings.

Photodetectors made of amorphous materials enable low-cost optical imaging and communications over non-semiconductor platforms. The key challenges are to improve efficiency, sensitivity, and frequency response. Using the localized surface plasmon resonance (LSPR) effect and an efficient carrier multiplication process, cycling excitation process (CEP), the plasmonically enhanced amorphous silicon photodetector (PEASP) with a thin (60 nm) absorption layer achieves a high external quantum efficiency with a record fast impulse response of 170 ps (FWHM). This approach offers the possibility of making detectors out of amorphous material for high frame rate imaging and optical communications in spite of the material’s low carrier mobility.

The COVID-19 has undergone several mutations, and caused deleterious effects on physical and mental health of people worldwide. Whilst physical exercise is known for its positive effect on enhancing immunity and reducing the negative consequences of unhealthy emotional states caused by the pandemic; there is a severe lack of psychological exercise intervention measures and mitigation strategies to advance the knowledge and role of physical exercise to improve mental health in most countries. This study surveyed the association between physical exercise and mental health burden during the COVID-19 outbreak in China to better understand the influence of different physical exercise types on reducing mental health burden during the pandemic. ANOVA, binary logistic regression, the chi-square test, and Spearman's correlation analysis were used for statistical analysis. 14,715 participants were included. The results showed that Chinese residents had several poor mental health conditions during the COVID-19 outbreak. And there was a significant positive correlation between the extent of adverse effects on mental health and provincial proportions of confirmed COVID-19 cases (r = 0.365, p < 0.05). Some main factors caused an unhealthy psychological status, including epidemic severity (62.77%, 95% CI 58.62-65.64%), prolonged home quarantine (60.84%, 95% CI 58.15-63.25%), spread of large amounts of negative information about COVID-19 in the media (50.78%, 95% CI 47.46-53.15%), limitations in daily life and social interaction (45.93%, 95%CI 42.46-47.55%), concerns about students' learning (43.13%, 95% CI 40.26-45.48%), and worries about being infected (41.13%, 95% CI 39.16-45.23%). There was a significant association between physical exercise and mental health. The largest associations were seen for home-based group entertainment exercise (i.e., family games, rope skipping, and badminton), Chinese traditional sports (i.e., Chinese martial arts, Taijiquan and Qigong), and popular sports (i.e., yoga, video dancing, sensory-motor games, and whole-body vibration), as well as durations of 30-60 min per session, frequencies of three to five times per week and a total of 120-270 min of moderate-intensity exercise weekly during the COVID-19 outbreak (p < 0.05).

High-order interactive features capture the correlation between different columns and thus are promising to enhance various learning tasks on ubiquitous tabular data. To automate the generation of interactive features, existing works either explicitly traverse the feature space or implicitly express the interactions via intermediate activations of some designed models. These two kinds of methods show that there is essentially a trade-off between feature interpretability and search efficiency. To possess both of their merits, we propose a novel method named Feature Interaction Via Edge Search (FIVES), which formulates the task of interactive feature generation as searching for edges on the defined feature graph. Specifically, we first present our theoretical evidence that motivates us to search for useful interactive features with increasing order. Then we instantiate this search strategy by optimizing both a dedicated graph neural network (GNN) and the adjacency tensor associated with the defined feature graph. In this way, the proposed FIVES method simplifies the time-consuming traversal as a typical training course of GNN and enables explicit feature generation according to the learned adjacency tensor. Experimental results on both benchmark and real-world datasets show the advantages of FIVES over several state-of-the-art methods. Moreover, the interactive features identified by FIVES are deployed on the recommender system of Taobao, a worldwide leading e-commerce platform. Results of an online A/B testing further verify the effectiveness of the proposed method FIVES, and we further provide FIVES as AI utilities for the customers of Alibaba Cloud., Comment: Accepted by KDD-21

In cardiac myocytes in vitro, hydrogen peroxide induces autophagic cell death and necroptosis. Oxidative stress, myocyte autophagy and necroptosis coexist in heart failure (HF). In this study, we tested the hypothesis that excessive oxidative stress mediates pathological autophagy and necroptosis in myocytes in pressure overload-induced HF. HF was produced by chronic pressure overload induced by abdominal aortic constriction (AAC) in rats. Rats with AAC or sham operation were randomised to orally receive an antioxidant N-acetylcysteine (NAC) or placebo for 4 weeks. Echocardiography was performed for the assessments of left ventricular (LV) structure and function. AAC rats exhibited decreased LV fractional shortening (FS) at 4 weeks after surgery. NAC treatment attenuated decreased LV FS in AAC rats. In AAC rats, myocardial level of 8-hydroxydeoxyguanosine assessed by immunohistochemical staining, indicative of oxidative stress, was increased, LC3 II protein, a marker of autophagy, Beclin1 protein and Atg4b, Atg5, Atg7 and Atg12 mRNA expression were markedly increased, RIP1, RIP3 and MLKL expression, indicative of necroptosis, was increased, and all of the alterations in AAC rats were prevented by the NAC treatment. NAC treatment also attenuated myocyte cross-sectional area and myocardial fibrosis in AAC rats. In conclusion, NAC treatment prevented the increases in oxidative stress, myocyte autophagy and necroptosis and the decrease in LV systolic function in pressure overload-induced HF. These findings suggest that enhanced oxidative stress mediates pathological autophagy and necroptosis in myocytes, leading to LV systolic dysfunction, and antioxidants may be of value to prevent HF through the inhibition of excessive autophagy and necroptosis.

Mining from graph-structured data is an integral component of graph data management. A recent trending technique, graph convolutional network (GCN), has gained momentum in the graph mining field, and plays an essential part in numerous graph-related tasks. Although the emerging GCN optimization techniques bring improvements to specific scenarios, they perform diversely in different applications and introduce many trial-and-error costs for practitioners. Moreover, existing GCN models often suffer from oversmoothing problem. Besides, the entanglement of various graph patterns could lead to non-robustness and harm the final performance of GCNs. In this work, we propose a simple yet efficient graph decomposition approach to improve the performance of general graph neural networks. We first empirically study existing graph decomposition methods and propose an automatic connectivity-ware graph decomposition algorithm, DeGNN. To provide a theoretical explanation, we then characterize GCN from the information-theoretic perspective and show that under certain conditions, the mutual information between the output after l layers and the input of GCN converges to 0 exponentially with respect to l. On the other hand, we show that graph decomposition can potentially weaken the condition of such convergence rate, alleviating the information loss when GCN becomes deeper. Extensive experiments on various academic benchmarks and real-world production datasets demonstrate that graph decomposition generally boosts the performance of GNN models. Moreover, our proposed solution DeGNN achieves state-of-the-art performances on almost all these tasks.

Background: Despite substantial improvement in chronic total occlusions (CTO) revascularization technique, the long-term clinical outcomes in diabetic patients with revascularized CTO remain controversial. Our study aimed to investigate the 5-year cardiovascular survival for patients with or without type 2 diabetes mellitus (DM) who underwent successful percutaneous coronary intervention (PCI) for CTO.Methods: Data of the current analysis derived from a large single-center, prospective and observational cohort study, including 10,724 patients who underwent PCI in 2013 at Fuwai Hospital. Baseline, angiographic and follow-up data were collected. The primary endpoint was major adverse cardiac and cerebrovascular events (MACCE), which consisted of death, recurrent myocardial infarction (MI), stroke and target vessel revascularization (TVR). The secondary endpoint was all-cause mortality. Cox regression analysis and propensity-score matching was performed to balance the baseline confounders.Results: A total of 719 consecutive patients with ≥1 successful CTO-PCI were stratified into diabetic (n = 316, 43.9%) and non-diabetic (n = 403, 56.1%) group. During a median follow-up of 5 years, the risk of MACCE (adjusted hazard ratio [HR] 1.47, 95% confidence interval [CI] 1.08–2.00, P = 0.013) was significantly higher in the diabetic group than in the non-diabetic group, whereas the adjusted risk of all-cause mortality (HR 2.37, 95% CI 0.94–5.98, P = 0.068) was similar. In the propensity score matched population, there were no significant differences in the risk of MACCE (HR 1.27, 95% CI 0.92–1.75, P = 0.155) and all-cause mortality (HR 2.56, 95% CI 0.91–7.24, P = 0.076) between groups. Subgroup analysis and stratification analysis revealed consistent effects on 5-year MACCE across various subgroups.Conclusions: In patients who received successful CTO-PCI, non-diabetic patients were related to better long-term survival benefit in terms of MACCE. The risk of 5-year MACCE appeared to be similar in less-controlled and controlled diabetic patients after successful recanalization of CTO. Further randomized studies are warranted to confirm these findings.

Background: There is controversy over the relationship between bilirubin and coronary artery disease. This study aimed to evaluate the predictive value of direct bilirubin (DB) in patients with complex acute coronary syndrome (ACS).Methods: From April 2004 to February 2011, 5,322 ACS patients presenting with three-vessel disease were consecutively enrolled. Disease severity and complexity were determined by SYNTAX score (SS) and SS II. The primary endpoint was all-cause death, and the secondary endpoints were major adverse cardiovascular and cerebrovascular events (MACCE). Stratification of normal glucose regulation, prediabetes, and diabetes was based on a previous diagnosis, hypoglycemic medications, fasting blood glucose, and hemoglobin A1c.Results: Subjects were divided into quartiles according to baseline DB (μmol/L): Q1 (0–1.6), Q2 (1.61–2.20), Q3 (2.21–2.80), and Q4 (>2.80). Multivariable logistic regression analysis showed that DB was an independent predictor of intermediate–high SS. During a median follow-up time of 6.5 years, elevated DB was associated with more all-cause death (p < 0.001) but not MACCE. DB remained to be predictive of all-cause death in the multivariable Cox regression model (Q2 vs. Q1: HR 1.043, 95% CI 0.829–1.312, p = 0.719; Q3 vs. Q1: HR 1.248, 95% CI 1.001–1.155, p = 0.048; Q4 vs. Q1: HR 1.312, 95% CI 1.063–1.620, p = 0.011). When subjects are stratified according to glucose metabolism regulation and treatment strategies, the predictivity of DB was only profound in patients with diabetes or with conservative treatment. Additionally, incorporating DB further improved the discrimination and reclassification abilities of SS II for risk prediction.Conclusion: DB is a potential biomarker for predicting lesion severity and long-term outcomes in ACS patients.

This paper proposes an automatic method for detection and mapping of Espeletia plants from aerial images acquired by UAV drone. The proposed approach integrated a computer vision for automatic extraction of training zones and tested on three well-established machine learning algorithms to detect regions belonging to Espeletia plants. The main components of the method are: (i) data capture and preprocessing; (ii) automatic extraction of training zones; and (iii) classification procedure using machine learning algorithms. Experimental results show that the method can achieve accurate detection and mapping of Espeletia plants, with up to 98.3% accuracy.

Cloud removal in optical remote sensing imagery is essential for many Earth observation applications. To recover the cloud obscured information, some preconditions must be satisfied. For example, the cloud must be semitransparent or relationships between contaminated and cloud-free pixels must be assumed. Due to the inherent imaging geometry features in satellite remote sensing, it is impossible to observe the ground under the clouds directly; therefore, cloud removal algorithms are always not perfect owing to the loss of ground truth. Recently, the use of passenger aircraft as a platform for remote sensing has been proposed by some researchers and institutes, including Airbus and the Japan Aerospace Exploration Agency. Passenger aircraft have the advantages of short visitation frequency and low cost. Additionally, because passenger aircraft fly at lower altitudes compared to satellites, they can observe the ground under the clouds at an oblique viewing angle. In this study, we examine the possibility of creating cloud-free remote sensing data by stacking multiangle images captured by passenger aircraft. To accomplish this, a processing framework is proposed, which includes four main steps: first, multiangle image acquisition from passenger aircraft, second, cloud detection based on deep learning semantic segmentation models, third, cloud removal by image stacking, and fourth, image quality enhancement via haze removal. This method is intended to remove cloud contamination without the requirements of reference images and predetermination of cloud types. The proposed method was tested in multiple case studies, wherein the resultant cloud- and haze-free orthophotos were visualized and quantitatively analyzed in various land cover type scenes. The results of the case studies demonstrated that the proposed method could generate high quality, cloud-free orthophotos. Therefore, we conclude that this framework has great potential for creating cloud-free remote sensing images when the cloud removal of satellite imagery is difficult or inaccurate.

BackgroundIn euthyroid patients undergoing percutaneous coronary intervention (PCI), it is still unclear whether free triiodothyronine to free thyroxine (FT3/FT4) ratio can predict the recurrence of cardiovascular events (CVEs). We aim to investigate its association with recurrent long-term adverse events in this population.Methods3549 euthyroid patients with prior CVEs history undergoing PCI were consecutively enrolled in our study and subsequently divided into three FT3/FT4 ratio tertiles (T12.75, n=1181). The primary endpoint was major adverse cardiovascular and cerebrovascular event (MACCE), a composite of all-cause death, myocardial infarction, stroke and revascularization. The secondary endpoints were all-cause death and cardiac death.ResultsThe median follow-up time was 5 years. The incidence of all-cause death, cardiac death and MACCE were significantly higher among patients in the lowest FT3/FT4 tertile (PConclusionsIn euthyroid patients with prior cardiovascular events undergoing PCI, FT3/FT4 ratio might be a potential predictor of all-cause and cardiac mortality. Routine assessment of FT3/FT4 ratio might be a simple and effective tool for risk stratification in this specific patient population.

The capsule network (Caps) is a novel type of neural network that has great potential for the classification of hyperspectral remote sensing. However, the Caps suffers from the issue of gradient vanishing. To solve this problem, a powered activation regularization based adaptive capsule network (PAR-ACaps) was proposed for hyperspectral remote sensing classification, in which an adaptive routing algorithm without iteration was applied to amplify the gradient, and the powered activation regularization method was used to learn the sparser and more discriminative representation. The classification performance of PAR-ACaps was evaluated using two public hyperspectral remote sensing datasets, i.e., the Pavia University (PU) and Salinas (SA) datasets. The average overall classification accuracy (OA) of PAR-ACaps with shallower architecture was measured and compared with those of the benchmarks, including random forest (RF), support vector machine (SVM), 1-dimensional convolutional neural network (1DCNN), two-dimensional convolutional neural network (CNN), three-dimensional convolutional neural network (3DCNN), Caps, and the original adaptive capsule network (ACaps) with comparable network architectures. The OA of PAR-ACaps for PU and SA datasets was 99.51% and 94.52%, respectively, which was higher than those of benchmarks. Moreover, the classification performance of PAR-ACaps with relatively deeper neural architecture (four and six convolutional layers in the feature extraction stage) was also evaluated to demonstrate the effectiveness of gradient amplification. As shown in the experimental results, the classification performance of PAR-ACaps with relatively deeper neural architecture for PU and SA datasets was also superior to 1DCNN, CNN, 3DCNN, Caps, and ACaps with comparable neural architectures. Additionally, the training time consumed by PAR-ACaps was significantly lower than that of Caps. The proposed PAR-ACaps is, therefore, recommended as an effective alternative for hyperspectral remote sensing classification.

Recent advances in machine learning (ML) systems have made it incredibly easier to train ML models given a training set. However, our understanding of the behavior of the model training process has not been improving at the same pace. Consequently, a number of key questions remain: How can we systematically assign importance or value to training data with respect to the utility of the trained models, may it be accuracy, fairness, or robustness? How does noise in the training data, either injected by noisy data acquisition processes or adversarial parties, have an impact on the trained models? How can we find the right data that can be cleaned and labeled to improve the utility of the trained models? Just when we start to understand these important questions for ML models in isolation recently, we now have to face the reality that most real-world ML applications are way more complex than a single ML model. In this article---an extended abstract for an invited talk at the DEEM workshop---we will discuss our current efforts in revisiting these questions for an end-to-end ML pipeline, which consists of a noise model for data and a feature extraction pipeline, followed by the training of an ML model. In our opinion, this poses a unique challenge on the joint analysis of data processing and learning. Although we will describe some of our recent results towards understanding this interesting problem, this article is more of a "confession" on our technical struggles and a "cry for help" to our data management community.

The prognostic value of high-sensitivity C-reactive protein (hsCRP) in complex coronary artery disease has not been fully established. We aimed to determine the association between hsCRP and long-term outcomes in elderly patients with 3-vessel disease (TVD). From April 2004 to February 2011, 3069 patients aged ≥65 years with TVD were consecutively enrolled and received medical treatment alone, percutaneous coronary intervention, or coronary artery bypass grafting. The patients were divided into 2 groups according to their hsCRP levels

Background There are relatively limited data regarding real-world outcomes in very old patients with three-vessel disease (3VD) receiving different therapeutic strategies. This study aimed to perform analysis of long-term clinical outcomes of medical therapy (MT), coronary artery bypass grafting (CABG), and percutaneous coronary intervention (PCI) in this population. Methods We included 711 patients aged ≥ 75 years from a prospective cohort of patients with 3VD. Consecutive enrollment of these patients began from April 2004 to February 2011 at Fu Wai Hospital. Patients were categorized into three groups (MT, n = 296; CABG, n = 129; PCI, n = 286) on the basis of different treatment strategies. Results During a median follow-up of 7.25 years, 262 deaths and 354 major adverse cardiac and cerebrovascular events (MACCE) occurred. Multivariate Cox analysis showed that the risk of cardiac death was significantly lower for CABG compared with PCI (adjusted hazard ratio [HR] = 0.475, 95% confidence interval [CI] 0.232–0.974, P = 0.042). Additionally, MACCE appeared to show a trend towards a better outcome for CABG (adjusted HR = 0.759, 95% CI 0.536–1.074, P = 0.119). Furthermore, CABG was significantly superior in terms of unplanned revascularization (adjusted HR = 0.279, 95% CI 0.079–0.982, P = 0.047) and myocardial infarction (adjusted HR = 0.196, 95% CI 0.043–0.892, P = 0.035). No significant difference in all-cause death between CABG and PCI was observed. MT had a higher risk of cardiac death than PCI (adjusted HR = 1.636, 95% CI 1.092–2.449, P = 0.017). Subgroup analysis showed that there was a significant interaction between treatment strategy (PCI vs. CABG) and sex for MACCE (P = 0.026), with a lower risk in men for CABG compared with that of PCI, but not in women. Conclusions CABG can be performed with reasonable results in very old patients with 3VD. Sex should be taken into consideration in therapeutic decision-making in this population.

Quantifying the importance of each training point to a learning task is a fundamental problem in machine learning and the estimated importance scores have been leveraged to guide a range of data workflows such as data summarization and domain adaption. One simple idea is to use the leave-one-out error of each training point to indicate its importance. Recent work has also proposed to use the Shapley value, as it defines a unique value distribution scheme that satisfies a set of appealing properties. However, calculating Shapley values is often expensive, which limits its applicability in real-world applications at scale. Multiple heuristics to improve the scalability of calculating Shapley values have been proposed recently, with the potential risk of compromising their utility in real-world applications. \textit{How well do existing data quantification methods perform on existing workflows? How do these methods compare with each other, empirically and theoretically? Must we sacrifice scalability for the utility in these workflows when using these methods?} In this paper, we conduct a novel theoretical analysis comparing the utility of different importance quantification methods, and report extensive experimental studies on existing and proposed workflows such as noisy label detection, watermark removal, data summarization, data acquisition, and domain adaptation. We show that Shapley value approximation based on a $K$NN surrogate over pre-trained feature embeddings obtains comparable utility with existing algorithms while achieving significant scalability improvement, often by orders of magnitude. Our theoretical analysis also justifies its advantage over the leave-one-out error. The code is available at \url{https://github.com/AI-secure/Shapley-Study}.

Aerospace milestones in human history, including returning to the moon and manned Martian missions, have been implemented in recent years. Space exploration has become one of the global common goals, and to ensure the survival and development of human beings in the extraterrestrial extreme environment has been becoming the basic ability and technology of manned space exploration. For the purpose of fulfilling the goal of extraterrestrial survival, researchers in Nanjing University and the China Academy of Space Technology proposed extraterrestrial artificial photosynthesis (EAP) technology. By simulating the natural photosynthesis of green plants on the Earth, EAP converts CO2/H2O into fuel and O2 in an in-situ, accelerated and controllable manner by using waste CO2 in the confined space of spacecraft, or abundant CO2 resources in extraterrestrial celestial environments, e.g. Mars. Thus, the material loading of manned spacecraft can be greatly reduced to support affordable and sustainable deep space exploration. In this paper, EAP technology is compared with existing methods of converting CO2/H2O into fuel and O2 in the aerospace field, especially the Sabatier method and Bosch reduction method. The research progress of possible EAP materials for in-situ utilization of extraterrestrial resources are also discussed in depth. Finally, this review lists the challenges that the EAP process may encounter, which need to be focused on for future implementation and application. We expect to deepen the understanding of artificial photosynthetic materials and technologies, and aim to strongly support the development of manned spaceflight., The extraterrestrial artificial photosynthesis in situ converts CO2 in spacecraft's confined space and in Martian atmosphere into O2 and fuels, thus supporting affordable and sustainable deep-space exploration and migration plans.

The appeal of serverless (FaaS) has triggered a growing interest on how to use it in data-intensive applications such as ETL, query processing, or machine learning (ML). Several systems exist for training large-scale ML models on top of serverless infrastructures (e.g., AWS Lambda) but with inconclusive results in terms of their performance and relative advantage over "serverful" infrastructures (IaaS). In this paper we present a systematic, comparative study of distributed ML training over FaaS and IaaS. We present a design space covering design choices such as optimization algorithms and synchronization protocols, and implement a platform, LambdaML, that enables a fair comparison between FaaS and IaaS. We present experimental results using LambdaML, and further develop an analytic model to capture cost/performance tradeoffs that must be considered when opting for a serverless infrastructure. Our results indicate that ML training pays off in serverless only for models with efficient (i.e., reduced) communication and that quickly converge. In general, FaaS can be much faster but it is never significantly cheaper than IaaS.

Blockchain has emerged as a promising solution for secure data storage and retrieval for decentralized applications. To scale blockchain systems, a prevailing approach is to employ a hybrid storage model, where only small meta-data are stored on-chain while the raw data are outsourced to an off-chain storage service provider. The key issue for query processing in such a system is the design of gas-efficient authenticated data structure (ADS) to authenticate the query results. In this paper, we study novel ADS schemes for authenticated keyword search in hybrid-storage blockchains. We first propose the Suppressed Merkle inverted (Merkleinv) index, which maintains only a partial ADS structure on-chain that can be securely updated with a logarithm-sized cryptographic proof. Moreover, we propose a Chameleon inverted (Chameleoninv) index that leverages the chameleon vector commitment to achieve a constant maintenance cost. It is further optimized with Bloom filters to enhance the query and verification performance. We prove the security of the proposed ADS schemes and evaluate their performance using real datasets on the Ethereum platform. Experimental results show that, compared to a baseline solution, the proposed Merkleinv and Chameleoninv indexes reduce the average on-chain maintenance cost from US$10.39 down to US$2.50 and US$0.24, respectively, without sacrificing much the query performance.

Background: Despite substantial improvement in chronic total occlusions (CTO) revascularization technique, the long-term clinical outcomes in diabetic patients with revascularized CTO remain controversial. Our study aimed to investigate the five-year cardiovascular survival for patients with or without type 2 diabetes mellitus (DM) who underwent successful percutaneous coronary intervention (PCI) for CTO. Methods: Data of the current analysis derived from a large single-center, prospective and observational cohort study, including 10,724 patients who underwent PCI in 2013 at Fuwai Hospital. Baseline, angiographic and follow-up data were collected. The primary endpoint was major adverse cardiac and cerebrovascular events (MACCE), which consisted of death, recurrent myocardial infarction (MI), stroke and target vessel revascularization (TVR). The secondary endpoint was all-cause mortality. Cox regression analysis and propensity-score matching was performed to balance the baseline confounders. Results: A total of 719 consecutive patients with ≥ 1 successful CTO-PCI were stratified into diabetic (n=316, 43.9%) and non-diabetic (n=403, 56.1%) group. During a median follow-up of 5 years, the risk of MACCE (adjusted hazard ratio [HR] 1.47, 95% confidence interval [CI] 1.08-2.00, P = 0.013) was significantly higher in the diabetic group than in the non-diabetic group, whereas the adjusted risk of all-cause mortality (HR 2.37, 95% CI 0.94-5.98, P = 0.068) was similar. In the propensity score matched population, there were no significant differences in the risk of MACCE (HR 1.27, 95% CI 0.92-1.75, P = 0.155) and all-cause mortality (HR 2.56, 95% CI 0.91-7.24, P = 0.076) between groups. Subgroup analysis revealed a consistent effect on five-year MACCE across various subgroups.Conclusions: In patients who received successful CTO-PCI, non-diabetic patients were related to better long-term survival benefit in terms of MACCE. Further randomized studies are warranted to confirm these findings.

Data quality affects machine learning (ML) model performances, and data scientists spend considerable amount of time on data cleaning before model training. However, to date, there does not exist a rigorous study on how exactly cleaning affects ML -- ML community usually focuses on developing ML algorithms that are robust to some particular noise types of certain distributions, while database (DB) community has been mostly studying the problem of data cleaning alone without considering how data is consumed by downstream ML analytics. We propose a CleanML study that systematically investigates the impact of data cleaning on ML classification tasks. The open-source and extensible CleanML study currently includes 14 real-world datasets with real errors, five common error types, seven different ML models, and multiple cleaning algorithms for each error type (including both commonly used algorithms in practice as well as state-of-the-art solutions in academic literature). We control the randomness in ML experiments using statistical hypothesis testing, and we also control false discovery rate in our experiments using the Benjamini-Yekutieli (BY) procedure. We analyze the results in a systematic way to derive many interesting and nontrivial observations. We also put forward multiple research directions for researchers., published in ICDE 2021