Search

The overwhelming parameters and computation consumption of deep neural networks limit their applicability to a single computing node with poor computing power, such as edge and mobile devices. Most previous works leverage model pruning and compression strategies to reduce DNN parameters for resource-constrained devices. However, most model compression methods may suffer from accuracy loss. Recently, we find that combine many weak computing nodes as a distributed system to run large and sophisticated DNN models is a promising solution for the issue. However, it is essential for the distributed system to design distributed DNN models and inference schemes, one of the great challenges of distributed system is how to design an efficient distributed DNN model for data parallelism and model parallelism, and communication overhead is also another critical performance bottleneck for distributed DNN model. Therefore, in this article, we propose DFSNet framework (Dividing-Fuse neural Network with Searching Strategy) for distributed DNN architecture. Firstly, the DFSNet framework includes a joint ”dividing-fusing” method to convert regular DNN models into distributed models that are friendly for distributed systems. This method divides the conventional DNN model in the channel dimension, and sets a few special layers to fuse feature-map information from different channel groups for accuracy improvement. Since the fusion layers are sparse in the network, they do not increase too much extra inference time and communication overhead on the distributed nodes, but they can maintain the accuracy of distributed neural networks significantly. Secondly, considering the architecture of distributed computing nodes, we propose a parallel fusion topology to improve the utilization of different computing nodes. Lastly, the popular weight-sharing neural architecture search (NAS) technique is leveraged to search the position of fusion layers in the distributed DNN model for high accuracy and finally generate an efficient distributed DNN model. Compared with the original network, our converted distributed DNN achieves better performance (e.g. 1.88% precision boosting in ResNet56 on CIFAR-100 dataset, and 1.25% precision improving in MobileNetV2 on ImageNet dataset). In addition, most layers of DNN have been divided into different distributed nodes on channel dimension, which is particularly suitable for distributed DNN architecture with very low communication overhead.

The past few years have witnessed the dramatic increase in layers of convolutional neural networks (CNN). Most studies focused on the CNN’s vertical structure design (e.g. residual structure, creating short paths architecture from early layers to later layers in vertical connections), but few people pay their attention to the process of feature generation and extraction in a single convolutional layer in CNN. In this paper, we find the non-feature suppression phenomenon in the process of extracting features. On the basis of this, we proposed an orthogonal approach named HSC (Horizontal Shortcut Connections) to improve feature representation fusion and computational efficiency for CNN. Especially, our HSC approach can effectively reduce interference overhead of non-feature areas and enhance the information fusion for depthwise convolution and group convolution which are the key blocks in lightweight neuron network. At HSC layer, the feature-maps of all preceding layer are properly connected with our strategy in horizon direction to constitute features and then produce a new representation which are used as input feature-maps passed on subsequent layers. Our HSC block can be plugged into convolution neural networks that include group convolution or depewise convolution, and can effectively improve accuracy of convolutional networks with slight additional computational cost. We evaluate our design on the popular lightweight neural networks and standard CNN structure. Compared with existing methods, we can achieve 1.63% accuracy improvement for MobileNet v2 on CIFAR-10 dataset and up to 3.70% accuracy improvement on CIFAR-100 dataset by adding HSC block after depthwise convolution, and 2.80% accuracy improvement on ImageNet dataset. For Mobilenet v3-small, we can achieve 0.8% accuracy improvement on ImageNet dataset. In order to prove the improvement effect of group convolution, the standard convolution is changed manually to group convolution and then the HSC block is added after group convolution, we can achieve 4X to 6X FLOPs improvement while maintaining the accuracy of neural networks. Notably, on ILSVRC- 2012, our method reduces more than 43% FLOPs on ResNet-50 without accuracy declines and reduces 60.1% FLOPs on ResNet-50 with 0.44% accuracy declines.We also present primary hardware experiment results when HSC framework running on special hardware platform.

Research and development (R&D)-intensive firms have strong incentives to maintain a smooth path for their R&D investments; otherwise, they will incur high adjustment costs. Examining data covering 2009 to 2016, we find that Chinese R&D-intensive firms, especially those with high innovation efficiency, tend to sell operating and financial assets to protect their value-enhancing R&D investments. However, financial constraints have adverse impacts on R&D smoothing with asset sales unless the firms have high innovation efficiency. The results suggest that innovation efficiency offers R&D-intensive firms, even financially constrained ones, a strong motivation to covert asset sales proceeds into R&D inputs, as the proceeds from asset sales provide a less-costly substitute for external financing. Given the importance of R&D for economic growth and the limited external financing opportunities in emerging capital markets like China, our findings reveal new insights regarding R&D financing.

Shadow prices of undesirable outputs assess the opportunity costs of emissions reduction and provide a comprehensive measuring indicator for the overall environmental regulation stringency. Shadow prices of industrial air pollutant emissions in China for 2006–2016 have been estimated by different nonparametric models, and their determinants, as well as the emissions controlling effects of environmental regulations, have been further investigated in this study. Our empirical results reveal that the conventional polluting technology model is consistent with the two sub-technologies model, and that shadow prices estimated by different models differed in absolute values but shared the same growth trends and regional characteristics. Decomposition results according to the logarithmic mean divisia index (LMDI) method indicate that variations in the shadow prices were due to the combined forces of the energy saving and end-of-pipe controlling effects in the analyzed periods. Moreover, we found that China’s environmental regulations with regional disparities in recent years (2006–2016) have effectively reduced industrial air pollutant emissions at an increasing rate, and that the environmental taxes imposed in 2018 still followed the existing regional unbalances of environmental regulation stringency.

Achieving carbon neutrality depends on carbon emission reduction and sequestration. However, research on vegetative carbon sequestration in China remains preliminary. In this study, we calculated the total carbon sequestration of terrestrial vegetation and that of different vegetation types in China from 2001–2019 using satellite data. Total vegetative carbon sequestration slowly increased but its increase was significantly lower than that of carbon emissions over the same period. Provinces with the strongest carbon sequestration capacity were mainly distributed in the south, whereas those with the lowest capacity were mainly in the west of Heihe-Tengchong Line. Woody grassland achieved the largest amount of carbon sequestration and grassland experienced the fastest growth. As sequestration ability varies by vegetation type and region, we suggest adopting a holistic regional approach that optimizes local vegetation growth environments, improves ecological compensation protection mechanisms (especially in urban areas) and develops other carbon sequestration pathways.

As fundamental data, gross domestic product (GDP) and electricity consumption can be used to effectively evaluate economic status and living standards of residents. Some scholars have estimated gridded GDP and electricity consumption. However, such gridded data have shortcomings, including overestimating real GDP growth, ignoring the heterogeneity of the spatiotemporal dynamics of the grid, and limited time-span. Simultaneously, the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP/OLS) and National Polar-orbiting Partnership’s Visible Infrared Imaging Radiometer (NPP/VIIRS) nighttime light data, adopted in these studies as a proxy tool, still facing shortcomings, such as imperfect matching results, discontinuity in temporal and spatial changes. In this study, we employed a series of methods, such as a particle swarm optimization-back propagation (PSO-BP) algorithm, to unify the scales of DMSP/OLS and NPP/VIIRS images and obtain continuous 1 km × 1 km gridded nighttime light data during 1992–2019. Subsequently, from a revised real growth perspective, we employed a top-down method to calculate global 1 km × 1 km gridded revised real GDP and electricity consumption during 1992–2019 based on our calibrated nighttime light data.

Pruning can remove redundant parameters and structures of Deep Neural Networks (DNNs) to reduce inference time and memory overhead. As an important component of neural networks, the feature map (FM) has stated to be adopted for network pruning. However, the majority of FM-based pruning methods do not fully investigate effective knowledge in the FM for pruning. In addition, it is challenging to design a robust pruning criterion with a small number of images and achieve parallel pruning due to the variability of FMs. In this paper, we propose Adaptive Knowledge Extraction for Channel Pruning (AKECP), which can compress the network fast and efficiently. In AKECP, we first investigate the characteristics of FMs and extract effective knowledge with an adaptive scheme. Secondly, we formulate the effective knowledge of FMs to measure the importance of corresponding network channels. Thirdly, thanks to the effective knowledge extraction, AKECP can efficiently and simultaneously prune all the layers with extremely few or even one image. Experimental results show that our method can compress various networks on different datasets without introducing additional constraints, and it has advanced the state-of-the-arts. Notably, for ResNet-110 on CIFAR-10, AKECP achieves 59.9% of parameters and 59.8% of FLOPs reduction with negligible accuracy loss. For ResNet-50 on ImageNet, AKECP saves 40.5% of memory footprint and reduces 44.1% of FLOPs with only 0.32% of Top-1 accuracy drop.

Firms in transition economies often suffer financial constraints. In Initial Public Offerings (IPOs), however, many newly listed Chinese firms raise funds in excess of what is originally planned. This paper examines whether the excess IPO funds are wasted on value-destroying spending or enable firms to take growth opportunities. After controlling for the endogeneity issue, we find that Chinese firms with excess IPO funds have better post-IPO operating performance, especially those with limited financing channels. In revealing the mechanism, we find that excess IPO fundraising alleviates financial constraints and reduces cost of debt.

Dental age estimation is widely used in forensic identification, but the accuracy of traditional methods cannot satisfy the demand for accuracy, especially for age estimation of adults. We introduce a deep learning-based methodology to estimate the age based on collected X-ray images of the teeth. We present a new dental dataset, which contains labeled orthopan-tomograms (OPGs) of 27,957 people, including 16,383 OPGs for females as well as 11,574 OPGs for males. All ages range from 0 to 93-year-old with a median of 27. The accuracy of the age labels is guaranteed by the ID card information. Aiming at the characteristics of the dental data itself, we explore various neural network elements that are effective for age estimation, including proper network depth, convolution kernel size, multi-branch structure, and the feature reusing of early layers. Based on the characteristic exploration, we further search models for dental age estimation by using the popular Neural Architecture Search (NAS) method. Experiment results show that our model achieves a mean absolute error (MAE) of 1.64 years, surpass all existing CNN models. Compared with Inception-v4 with an MAE of 1.70 and 20.46B FLOPs (inputs size 384×384), the FLOPs of our model can be reduced by 2.7 times (7.49B FLOPs). To our best knowledge, this is the first study for age estimation by exploring and searching the DNN model. Our results have surpassed legal medical expert-level performance (with an MAE of more than 2) for age estimation. Our methodology and results in this paper are very meaningful to forensic medicine for aging estimation with panoramic radiograph images.

Tumor cells actively release large quantities of exosomes, which pivotally participate in the regulation of cancer biology, including head and neck cancer (HNC). Exosome biogenesis and release are complex and elaborate processes that are considered to be similar to the process of exocyst-mediated vesicle delivery. By analyzing the expression of exocyst subunits and their role in patients with HNC, we aimed to identify exocyst and its functions in exosome biogenesis and investigate the molecular mechanisms underlying the regulation of exosome transport in HNC cells. We observed that exocysts were highly expressed in HNC cells and could promote exosome secretion in these cells. In addition, downregulation of exocyst expression inhibited HN4 cell proliferation by reducing exosome secretion. Interestingly, immunofluorescence and electron microscopy revealed the accumulation of multivesicular bodies (MVBs) after the knockdown of exocyst. Autophagy, the major pathway of exosome degradation, is not activated by this intracellular accumulation of MVBs, but these MVBs are consumed when autophagy is activated under the condition of cell starvation. Rab11a, a small GTPase that is involved in MVB fusion, also interacted with the exocyst. These findings suggest that the exocyst can regulate exosome biogenesis and participate in the malignant behavior of tumor cells., Graphical abstract, Exosomes play a role in the occurrence and development of head and neck cancer (HNC); however, the specific mechanism for their secretion remains unclear. We found that the multivesicular body gets transported to the plasma membrane for exosome secretion by exocyst via binding with Rab11a in HNC.

The transient receptor potential polycystin-2 (TRPP2) is encoded by the Pkd2 gene, and mutation of this gene can cause autosomal dominant polycystic kidney disease (ADPKD). Some patients with ADPKD experience extrarenal manifestations, including radiologic and clinical bronchiectasis. We hypothesized that TRPP2 may regulate airway smooth muscle (ASM) tension. Thus, we used smooth muscle- Pkd2 conditional knockout ( Pkd2SM-CKO) mice to investigate whether TRPP2 regulated ASM tension and whether TRPP2 deficiency contributed to bronchiectasis associated with ADPKD. Compared with wild-type mice, Pkd2SM-CKOmice breathed more shallowly and faster, and their cross-sectional area ratio of bronchi to accompanying pulmonary arteries was higher, suggesting that TRPP2 may regulate ASM tension and contribute to the occurrence of bronchiectasis in ADPKD. In a bioassay examining isolated tracheal ring tension, no significant difference was found for high-potassium-induced depolarization of the ASM between the two groups, indicating that TRPP2 does not regulate depolarization-induced ASM contraction. By contrast, carbachol-induced contraction of the ASM derived from Pkd2SM-CKOmice was significantly reduced compared with that in wild-type mice. In addition, relaxation of the carbachol-precontracted ASM by isoprenaline, a β-adrenergic receptor agonist that acts through the cAMP/adenylyl cyclase pathway, was also significantly attenuated in Pkd2SM-CKOmice compared with that in wild-type mice. Thus, TRPP2 deficiency suppressed both contraction and relaxation of the ASM. These results provide a potential target for regulating ASM tension and for developing therapeutic alternatives for some ADPKD complications of the respiratory system or for independent respiratory disease, especially bronchiectasis.

Based on the data of the BP Statistical Review of World Energy, this paper constructs the consumption and import–export of natural gas identities. It discusses the drivers of changes in global natural gas consumption and trade flows from 2008 to 2015 using the extended logarithmic mean Divisia index. The results show that differences in the natural gas supply and demand across countries or regions, as well as the distribution of energy between the domestic and international markets, can be better explained when natural gas trade movements are considered. By comparing the supply and consumption increment of natural gas, this study finds that only the energy intensity, economic growth, and demographic effects are consistent with each other. The changes in the impact of other effects mainly depend on storage variations and statistical errors. In addition, the primary drivers of the incremental changes in natural gas consumption vary in different countries. They include production scale, import scale, export scale, consumption structure proportion, energy intensity, economic growth, and population and balance effects. Finally, the consumption competitiveness of the liquefied natural gas significantly improved over the examined period.

China’s financial development and economic growth is achieved under weak legal institutions. The literature attributes this counter-example of law–finance–growth nexus to (a) alternative mechanisms in China such as incentives, reputation and relationships and(b) a well-functioning xinfang system with common law features.In recent years, China has made increasing efforts to strengthen its rule of law. The Communist Party of China (CPC) has taken the lead by launching a far-reaching campaign against corruption, establishing a system of inspection tours, and promulgating a large number of regulations. We argue that using regulations to complement laws is effective: CPC has enough bureaucratic prowess to crack down on corruption whereas the courts are subject to sub-version by powerful interests. We also discuss the drawbacks of this approach: regulations aiming at ex ante control of corruption substantially increase procedural formalism and limit the discretion of local governments and state-owned enterprises.

This study proposes a new approach to analyse the effects of an overlap term on the calculation of the overall Gini coefficient and estimates China’s Gini ratios since the adoption of the economic reform and open-door policies. A decomposition of the Chinese Gini coefficient for 1978–2010 reveals that the key factor contributing to income inequalities is the income disparity between rural and urban inhabitants. We further investigate the features of this income inequality between rural and urban areas and employ statistical approaches to evaluate the effects of urbanisation and rural-to-urban average income on nationwide income inequality. The results show that accelerating the pace of urbanisation is mainly responsible for decreasing China’s income disparity. Drawing on these results, we conclude with suggestions for related policies.

Eight years ago, The European Journal of Finance (EJF) published the first special issue on the Chinese Capital Markets. This was at a time when Chinese topics were not widely accepted by internationally excellent finance journals. Since then, The European Journal of Finance had published four special issues both containing and encouraging studies that reveal the unique institutions of China. This editorial summarises the articles included in this 5th CCM special issue and discusses a series of legislative and regulatory changes, which has strengthened the role of the Chinese Communist Party in the economy and society under the “Party leads everything” doctrine. We encourage future studies to examine their implications on finance and economics and believe that the results of such studies will be helpful for the CPC and the Chinese Government to shape future policies to ensure sustainable development.

There is a global consensus that carbon dioxide and other greenhouse gas emissions must be reduced as a response to global climate change. Remote sensing satellite data have become an important means of monitoring carbon emission due to its unique advantages such as availability, high resolution, and wide coverage, and remote sensing data are playing an increasingly important role in carbon monitoring and fixation. This article summarizes the main applications of remote sensing satellite data in reducing carbon emissions and prospects for future research.

IntroductionDiabetes-associated endothelial barrier function impairment might be linked to disturbances in Ca2+ homeostasis. To study the role and molecular mechanism of Orais–vascular endothelial (VE)-cadherin signaling complex and its downstream signaling pathway in diabetic endothelial injury using mouse aortic endothelial cells (MAECs).Research design and methodsThe activity of store-operated Ca2+ entry (SOCE) was detected by calcium imaging after 7 days of high-glucose (HG) or normal-glucose (NG) exposure, the expression levels of Orais after HG treatment was detected by western blot analysis. The effect of HG exposure on the expression of phosphorylated (p)-VE-cadherin and VE-cadherin on cell membrane was observed by immunofluorescence assay. HG-induced transendothelial electrical resistance was examined in vitro after MAECs were cultured in HG medium. FD-20 permeability was tested in monolayer aortic endothelial cells through transwell permeability assay. The interactions between Orais and VE-cadherin were detected by co-immunoprecipitation and immunofluorescence technologies. Immunohistochemical experiment was used to detect the expression changes of Orais, VE-cadherin and p-VE-cadherin in aortic endothelium of mice with diabetes.Results(1) The expression levels of Orais and activity of SOCE were significantly increased in MAECs cultured in HG for 7 days. (2) In MAECs cultured in HG for 7 days, the ratio of p-VE-cadherin to VE-cadherin expressed on the cell membrane and the FD-20 permeability in monolayer endothelial cells increased, indicating that intercellular permeability increased. (3) Orais and VE-cadherin can interact and enhance the interaction ratio through HG stimulation. (4) In MAECs cultured with HG, the SOCE activator ATP enhanced the expression level of p-VE-cadherin, and the SOCE inhibitor BTP2 decreased the expression level of p-VE-cadherin. (5) Significantly increased expression of p-VE-cadherin and Orais in the aortic endothelium of mice with diabetes.ConclusionHG exposure stimulated increased expression of Orais in endothelial cells, and increased VE-cadherin phosphorylation through Orais–VE-cadherin complex and a series of downstream signaling pathways, resulting in disruption of endothelial cell junctions and initiation of atherosclerosis.

Accurate, long-term, full-coverage carbon dioxide (CO2) data in units of prefecture-level cities are necessary for evaluations of CO2 emission reductions in China, which has become one of the world’s largest carbon-emitting countries. This study develops a novel method to match satellite-based Defense Meteorological Satellite Program’s Operational Landscan System (DMSP/OLS) and Suomi National Polar-orbiting Partnership’s Visible Infrared Imaging Radiometer Suite (NPP/VIIRS) nighttime light data, and estimates the CO2 emissions of 334 prefecture-level cities in China from 1992 to 2017. Results indicated that the eastern and coastal regions had higher carbon emissions, but their carbon intensity decreased more rapidly than other regions. Compared to previous studies, we provide the most extensive and long-term CO2 dataset to date, and these data will be of great value for further socioeconomic research. Specifically, this dataset provides a foundational data source for China’s future CO2 research and emission reduction strategies. Additionally, the methodology can be applied to other regions around the world.

This paper reviews the growing literature on initial coin offerings (ICOs) and provides original evidence of the investor protection on ICOs from 37 countries. We show that the anti-director rights and anti-self-dealing index are positively associated with the country-level raised fund of ICOs after controlling for economic and culture factors. The disclosure quality and investor rights as specified in the Whitepapers are generally poor and they are found to be important to raise more funds in ICOs. We argue that the lack of (self-) discipline poses a threat to investor protections. Around 60% Whitepaper do not disclose information on the use of proceeds or management team. Around 80% ICOs do not entitle investors the rights for dividend or vote. Our findings suggest the needs of regulating ICOs to protect investors.

Disease does not only affect development through its contemporaneous impact on health, but also through its enduring historical effect through its shaping of culture and institutions. By drawing on the experience of historical pandemics, we argue that some of the current stringent approaches adopted in combating Covid-19 come at the expense of social capital (trust) and institutions (checks on government power), which hold back long-run development. We review the evidence that the Black Death intensified witchcraft beliefs and antisemitism, which, in turn, developed mistrust and exerted an adverse influence on present-day development. Finally, we demonstrate that institutions and culture have greater explanatory power in terms of cross-country infection rates and fatality rates than does health care quality.

This paper examines the impact of historical endowment on the Initial Coin Offering (ICO) market. Leveraging a sample of 1,617 hand collected ICOs over 2015-2019 and regressions models, we find that a country's legal origin, disease environment, exposure to slave trade, and proximity to ancient trade routes are important determinants of the total amount of capital raised through ICOs. Further analyses suggest that the impact of historical endowment operates through legal and information sharing institutions. This finding suggests that traumatic shocks in the history plays an important role in shaping current Fintech development through long-run persistence of formal and informal institutions. This research highlights the importance to enlarge the scope of Fintech research and the importance to take institutions into account in relevant policy designs to promote financial development.