Search

Most deep neural networks (DNNs) consist fundamentally of convolutional and/or fully connected layers, wherein the linear transform can be cast as the product between a filter matrix and a data matrix obtained by arranging feature tensors into columns. Recently proposed deformable butterfly (DeBut) decomposes the filter matrix into generalized, butterfly-like factors, thus achieving network compression orthogonal to the traditional ways of pruning or low-rank decomposition. This work reveals an intimate link between DeBut and a systematic hierarchy of depthwise and pointwise convolutions, which explains the empirically good performance of DeBut layers. By developing an automated DeBut chain generator, we show for the first time the viability of homogenizing a DNN into all DeBut layers, thus achieving extreme sparsity and compression. Various examples and hardware benchmarks verify the advantages of All-DeBut networks. In particular, we show it is possible to compress a PointNet to <5% parameters with <5% accuracy drop, a record not achievable by other compression schemes.

Purpose: In response to the problem of insufficient traction/braking adhesion force caused by the existence of the third-body medium on the rail surface, this study aims to analyze the utilization of wheel-rail adhesion coefficient under different medium conditions and propose relevant measures for reasonable and optimized utilization of adhesion to ensure the traction/braking performance and operation safety of trains. Design/methodology/approach: Based on the PLS-160 wheel-rail adhesion simulation test rig, the study investigates the variation patterns of maximum utilized adhesion characteristics on the rail surface under different conditions of small creepage and large slip. Through statistical analysis of multiple sets of experimental data, the statistical distribution patterns of maximum utilized adhesion on the rail surface are obtained, and a method for analyzing wheel-rail adhesion redundancy based on normal distribution is proposed. The study analyzes the utilization of traction/braking adhesion, as well as adhesion redundancy, for different medium under small creepage and large slip conditions. Based on these findings, relevant measures for the reasonable and optimized utilization of adhesion are derived. Findings: When the third-body medium exists on the rail surface, the train should adopt the low-level service braking to avoid the braking skidding by extending the braking distance. Compared with the current adhesion control strategy of small creepage, adopting appropriate strategies to control the train’s adhesion coefficient near the second peak point of the adhesion coefficient-slip ratio curve in large slip can effectively improve the traction/braking adhesion redundancy and the upper limit of adhesion utilization, thereby ensuring the traction/braking performance and operation safety of the train. Originality/value: Most existing studies focus on the wheel-rail adhesion coefficient values and variation patterns under different medium conditions, without considering whether the rail surface with different medium can provide sufficient traction/braking utilized adhesion coefficient for the train. Therefore, there is a risk of traction overspeeding/braking skidding. This study analyzes whether the rail surface with different medium can provide sufficient traction/braking utilized adhesion coefficient for the train and whether there is redundancy. Based on these findings, relevant measures for the reasonable and optimized utilization of adhesion are derived to further ensure operation safety of the train.

In order to effectively convert greenhouse gas CO2into high value-added chemicals, simple and sustainable methods are urgently needed. In this study, a novel metal-anchored ionic porous organic polymer (M-iPOP) named Zn-HUT4 was synthesized via a very simple and low-cost synthesis process. Further investigation was conducted into the use of Zn-HUT4 as a heterogeneous catalyst in the cycloaddition reaction of CO2with epoxides. The obtained Zn-HUT4 catalyst had a rich porous structure and abundant metal anchor points, which can effectively promote mass transfer in the cycloaddition reaction and offer a substantial amount of Zn(II) ions as catalytic active sites. Using CO2/propylene oxide (PO) cycloaddition as a model reaction, the effect of a series of conditions on catalytic experiments was explored. The Zn-HUT4 catalyst had good recyclability and substrate universality.

To accelerate the inference of deep neural networks (DNNs), quantization with low-bitwidth numbers is actively researched. A prominent challenge is to quantize the DNN models into low-bitwidth numbers without significant accuracy degradation, especially at very low bitwidths $( &lt; 8$ bits). This work targets an adaptive data representation with variable-length encoding called DyBit. DyBit can dynamically adjust the precision and range of separate bit-fields to be adapted to the DNN weights/activations distribution. We also propose a hardware-aware quantization framework with a mixed-precision accelerator to tradeoff the inference accuracy and speedup. Experimental results demonstrate that the ImageNet inference accuracy via DyBit is 1.97% higher than the state-of-the-art at 4-bit quantization, and the proposed framework can achieve up to $8.1\times $ speedup compared with the original ResNet-50 model.

High-performance computing (HPC) can facilitate deep neural network (DNN) training and inference. Previous works have proposed multiple-precision floating- and fixed-point designs, but most can only handle either one independently. This brief proposes a novel reconfigurable processing element (PE) supporting both energy-efficient floating-point and fixed-point multiply-accumulate (MAC) operations. This PE can support $9\times $ BFloat16 (BF16), 4 $\times $ half-precision (FP16), $4\times $ TensorFloat-32 (TF32) and $1\times $ single-precision (FP32) MAC operation with 100% multiplication hardware utilization in one clock cycle. Besides, it can also support 72 $\times $ INT2, 36 $\times $ INT4 and 9 $\times $ INT8 dot product plus one 32-bit addend. The design is realized in a 28nm-process at a 1.471GHz slow-corner clock frequency. Compared with state-of-the-art (SOTA) multiple-precision PEs, the proposed work exhibits the best energy efficiency of 834.35GFLOPS/W and 1761.41GFLOPS/W at TF32 and BF16 with at least 10 $\times $ and 4 $\times $ improvement, respectively, for deep learning training. Meanwhile, this design supports energy-efficient fixed-point computing with a small hardware overhead for deep learning inference.

ABSTRACTThe limitation of human dental pulp stem cells (DPSCs), which have potential application value in regenerative medicine, is that they are prone to age in vitro. Studies have shown adrenomedullin (ADM) is believed to promote the proliferation of human DPSCs, but whether it can also affect aging remains to be investigated. A lentivirus vector was used to construct human DPSCs overexpressing ADM. Senescence tests were carried out on cells of the 7th and 15th passage. Transcriptome analysis was conducted to analyze microRNA expression regulation changes after human DPSCs overexpressed ADM. H2O2induced the aging model of human DPSCs, and we examined the mechanism of recovery of aging through transfection experiments with miR-152 mimic, pCDH-CCNA2, and CCNA2 siRNA. Overexpression of ADM significantly upregulated the G2/M phase ratio of human DPSCs in natural passage culture (P = 0.001) and inhibited the expression of p53 (P = 0.014), P21 WAF1(P = 0.015), and P16 INK4A(P = 0.001). Decreased ROS accumulation was observed in human DPSCs during long-term natural passage (P = 0.022). Transcriptome analysis showed that miR-152 was significantly upregulated during human DPSC senescence (P = 0.001) and could induce cell senescence by directly targeting CCNA2. Transfection with miR-152 mimic significantly reversed the inhibitory effect of ADM overexpression on p53 (P = 0.006), P21 WAF1(P = 0.012), and P16 INK4A(P = 0.01) proteins in human DPSCs (H2O2-induced). In contrast, pCDH-CCNA2 weakened the effect of the miR-152 mimic, thus promoting cell proliferation and antiaging. ADM-overexpressing human DPSCs promote cell cycle progression and resist cellular senescence through CCNA2 expression promotion by inhibiting miR-152.

In the face of the decline of time-honored brands, activating time-honored brands by means of product extension is an effective guarantee for the long-term development of time-honored brands and an important way for them to achieve lasting success. In this paper, Ye Chun Tea House is taken as the parent brand. Through consumer testing sashimi buns and hamburger are identified as two extension products with significant differentiation degree of perceived fit with the parent brand. Student’s t test and multiple regression analysis were used to explore the positive or negative impact of extension products with different perceived fit on brand, as well as the effect of perceived fit on brand attachment and brand loyalty. The cross-multiplicative model is used to discuss the strength of the above influence in different gender, age and income consumer groups. The empirical analysis and research conclusions of this paper focus on two aspects: first, extension products with high perceived fit improve brand loyalty, while extension products with low perceived fit reduce brand loyalty. Second, for extension products with high perceived fit, perceived fit positively affects brand loyalty. Brand attachment plays a mediating role between perceived fit and brand loyalty. Age had a negative moderating effect on perceived fit and brand attachment, perceived fit and brand loyalty, while income level only had a positive moderating effect on cultural fit and brand attachment, cultural fit and brand loyalty. Based on the research conclusions, this paper provides policy-making advice for government departments in four aspects: research and development of new products, optimization of financing services, loose loan policies and construction of omni-channel marketing platform.

SummaryThree new genera and species of cave-adapted trechine beetles are established and described from limestone caves in southwestern China: Yunaphaenops graciliphallusn. gen., n. sp.(Fengwei Dong Cave, Qiubei County, Yunnan Province), a semi-aphaenopsian species with its taxonomic position remaining unclear within the tribe Trechini though allied to SinaphaenopsUéno & Wang 1991 based on molecular analysis; Leshanius yunzhiin. gen., n. sp.(Xiaoxuankong Tiankeng Cave, Leshan Shi, Sichuan Province), an aphaenopsian species somewhat similar to the genus BoreaphaenopsUéno 2002 in morphology, but allied to SichuanotrechusDeuve, 2005 in molecular phylogeny; and Panotrechus pilosusn. gen., n. sp.(Wojiao Dong Cave, Panzhou, Guizhou Province), a peculiar genus with several remarkable characters in morphology, but allied to CkacusTian & Huang, 2021 in molecular analysis. Jujiroa duqianaen. sp.is also reported from Xiaoxuankong Tiankeng, allied to J. deliciolaUéno & Kishimoto, 2001. A new locality for J. deliciolais recorded. Finally, an updated key to all Sichuanese species of Jujiroais provided.

Community detection, which focuses on clustering vertex interactions, plays a significant role in network analysis. However, it also faces numerous challenges like missing data and adversarial attack. How to further improve the performance and robustness of community detection for real-world networks has raised great concerns. In this paper, we explore robust community detection by enhancing network structure, with two generic algorithms presented: one is named robust community detection via genetic algorithm (RobustECD-GA), in which the modularity and the number of clusters are combined in a fitness function to find the optimal structure enhancement scheme; the other is called robust community detection via similarity ensemble (RobustECD-SE), integrating multiple information of community structures captured by various vertex similarities, which scales well on large-scale networks. Comprehensive experiments on real-world networks demonstrate, by comparing with two traditional enhancement strategies, that the new methods help six representative community detection algorithms achieve more significant performance improvement. Moreover, experiments on the corresponding adversarial networks indicate that the new methods could also optimize the network structure to a certain extent, achieving stronger robustness against adversarial attack.

In this Letter, we analytically model the impact of polarization crosstalk in the polarization-multiplexed carrier self-homodyne (PMC-SH) system with adaptive polarization control technology. When the optical paths of the signal and local oscillator (LO) are matched well, it is found that the polarization crosstalk results in a nonlinear shift on the constellation. Thus, we further propose a compensation scheme based on a low-complexity polynomial nonlinear equalizer (PNLE). Both simulation and experimental results validate our theoretical analysis. Moreover, the proposed PNLE-based compensation scheme achieves up to 1.23?dB tolerance improvement with respect to polarization crosstalk for 20 Gbaud 64QAM in the experiment.

Thailand is the largest nature rubber exporter in the world. The rubber industry is particularly important for both the national economy and farmers. However, the increasing popularity of monoculture rubber plantation has posed challenges for the sustainable development of rubber industry due to the negative impacts of homogeneous cropping systems. Rubber crop diversity system is expected to alleviate the environment degradation and secure rubber farmers’ income. Nevertheless, the rubber diverse cropping systems adoption is more complicated for most rubber farmers in Thailand. Using the data on rubber farmers which cover 3344 observations from Office of Agricultural Economics in Thailand throughout the period of 2012–2014, this paper investigated the influential factors of rubber crop diversity adoption. The results indicate that rubber farmers are usually lack of management skills, information, and market access in crop diversity adoption. Adequate rainfall and water, household labour, land right certificate, land degradation, the use of organic fertilizer, crop sales channels and irrigation access have positive impacts on the adoption of rubber crop diversity. On the other hand, in the southern region, the hiring of foreign labours, agricultural training, off-farm work, male household heads as well as their age and education level are negatively associated with crop diversity adoption. According to the empirical results, it is necessary for rubber institutions to pay more attention to training programmes and education on rubber farmers’ diverse cropping system management skills. Moreover, rubber sustainable development strategies should offer rubber households additional incentives and financial support to relieve their financial constraints and improve the market accessibility. In addition, it is also necessary to focus on rubber households with poor natural resources and agricultural infrastructure and offer supports for infrastructure, such as irrigation systems, to increase rubber crop diversity adoption in Thailand.

Aerosol-transmitted viruses possess strong infectivity and can spread over long distances, earning the difficult-to-control title. They cause various human diseases and pose serious threats to human health. Mutations can increase the transmissibility and virulence of the strains, reducing the protection provided by vaccines and weakening the efficacy of antiviral drugs. In this study, we established a manually curated database (termed AVM) to store information on aerosol-transmitted viral mutations (VMs). The current version of the AVM contains 42,041 VMs (including 2613 immune escape mutations), 45 clinical information datasets, and 407 drugs/antibodies/vaccines. Additionally, we recorded 88 human diseases associated with viruses and found that the same virus can target multiple organs in the body, leading to diverse diseases. Furthermore, the AVM database offers a straightforward user interface for browsing, retrieving, and downloading information. This database is a comprehensive resource that can provide timely and valuable information on the transmission, treatment, and diseases caused by aerosol-transmitted viruses (http://www.bio-bigdata.center/AVM).

In this study, we isolated lovastatin derivatives, including aculeatiols A–G (1–7) and three known compounds (8–10), from Aspergillus aculeatus. Their structures and absolute configurations were experimentally determined by high-resolution electrospray ionization mass spectrometry, nuclear magnetic resonance spectroscopy, and X-ray diffraction analyses, and the results were corroborated by quantum-chemical calculations. As members of the lovastatin derivatives, aculeatiols A–C (1–3) possess a γ-lactone functional group in the side chain. Compound 6represents the first example that features an undescribed aromatized heterotetracyclic 6/6/6/6 ring system. Biologically, the lipid-lowering effects of all of these compounds were evaluated by analyzing the free fatty acid-induced intracellular lipid accumulation. In addition, compound 5, which regulated the transcription of genes associated with lipid uptake and synthesis, inhibited the accumulation of lipids.

An aminic CO2-based poly(ether-carbonate) polyol has been successfully synthesized by CO2/propylene oxide copolymerization with the primary aromatic amine as the initiator under a double metal cyanide (DMC) complex catalyst. The weak basicity and coordination ability of amine initiators are conducive to enhancing the copolymerization productivity and catalytic activity. A branched aminic CO2-polyol has been constructed with catalytic efficiency up to 4.1 kg g–1DMC using the diamine of 4,4′-methylene-bis(2-chloroaniline) as the initiator, whose average −OH functionality can be facilely tuned in the range of 2.4–3.6. The evolution of the number and length of multiarm polymers was investigated by nuclear magnetic resonance (NMR) and electrospray ionization-mass spectrometry (ESI-MS) analyses. A plausible mechanism of DMC catalyst-mediated copolymerization has also been proposed. Interestingly, the obtained low-molecular-weight (MW) aminic CO2-polyols displayed an intrinsic self-catalytic feature that is suitable for the preparation of rigid polyurethane foams without any external catalyst. This work demonstrates an efficient pathway to fix CO2into much more environmentally friendly polyols with low-volatility organic compounds.

Supplemental Digital Content is available in the text

Kidney cancer is a significant health concern with growing treatment resistance, often linked to genomic instability. This study used datasets from 72 renal and 952 clear cell renal cell carcinoma samples to identify genomic instability-derived lncRNAs and develop a prognostic index (GILPI).

Although supercapacitors are regarded as the next-generation power resources for various electronic devices, their broad-scale commercialization is largely dependent on the exploration and utilization of advanced electrode materials. Herein, we demonstrate an in situoxidation approach to the construction of 2D/2D heterojunctions built from α-MnO2nanolamellas grown on Ti3C2TxMXene nanosheets through an ambient-temperature redox reaction. Interestingly, the growth density of MnO2nanolamellas on MXene surface can be well controlled by regulating the reaction times, which endows the heterojunctions with adjustable composition, morphology, and microstructure. By virtue of the unique architectural merits including large accessible surface areas, uniform dispersion of MnO2nanolamellas, well-contacted 2D/2D interfaces, and high electrical conductivity, the optimized MnO2/MXene heterojunction synthesized at a reaction time of 36 h exhibits boosted electrochemical performance with a high specific capacitance of 334.3 F g−1at 0.5 A g−1, good rate property, and long cycle life with ~90.9 % capacitance retained after 5000 cycles, far outperforming those of bare MnO2and MXene electrodes.

To investigate the effect of nutritional factors on bone mineral density (BMD) using quantitative computed tomography (QCT) combined with blood biochemistry in patients on maintenance hemodialysis (MHD).

Modeling an accurate and effective heat source is a challenge when describing the thermal behavior of wheel/rail contact, conventionally modeled using Goldak’s heat source. This Gaussian distributed model transplanted directly from the welding field is quite different from the ellipsoidal heat source on the contact patch. An innovative heat source model which reflects the shape and size of contact patch was created based on the energy equivalence in this study. The thermal behavior of the wheel/rail contact was described by using this heat source moving along the rail. Numerical simulations based on a three-dimensional (3-D) finite element (FE) model validate that the proposed heat source model is valid as the relative errors of the peak temperature between the proposed and previously published heat source model are less than 4.45%. The distribution of the contact temperature on the rail surface reflects the shape and size of the contact patch. The semi-ellipsoidal distributed temperature obtained by the proposed elliptic cylinder model is more precise than the Gaussian distributed temperature obtained by the conventionally used Goldak’s model. The proposed elliptic cylinder model has the advantages of simple structure and flexible model variables than the ready-made Goldak’s model in commercial software.