Search

Transformer models have achieved remarkable results in various natural language tasks, but they are often prohibitively large, requiring massive memories and computational resources. To reduce the size and complexity of these models, we propose LoSparse (Low-Rank and Sparse approximation), a novel model compression technique that approximates a weight matrix by the sum of a low-rank matrix and a sparse matrix. Our method combines the advantages of both low-rank approximations and pruning, while avoiding their limitations. Low-rank approximation compresses the coherent and expressive parts in neurons, while pruning removes the incoherent and non-expressive parts in neurons. Pruning enhances the diversity of low-rank approximations, and low-rank approximation prevents pruning from losing too many expressive neurons. We evaluate our method on natural language understanding, question answering, and natural language generation tasks. We show that it significantly outperforms existing compression methods.

The origin and spread of the Chinese batik have always been a controversial issue in the field of Chinese dyeing and waving historiography. The systemic analysis was carried out by the methods of textile archaeology and literature research in this paper. Conclusions have been made as follows: there were two models in the origin of Chinese batik. One, such as the Central Plains area, was the type of external afference. Another model, such as the south-western ethnic areas, was the type of generating from the inside of the area. The transmutation of batik in Central Plain areas have passed through three stages: introduction from the Eastern Han Dynasty to the Northern and Southern Dynasties, mass acceptance during the Sui Dynasty, Tang Dynasty and Five Dynasties, obsolescence during the Two Song Dynasties. From the perspectives of the reason, the rapid disappearance of batik in the Central Plains areas were caused by the special internal and external environment and the development trend of dyeing technology of the Song Dynasty.

As a parent compound of Li-rich electrodes, Li2MnO3 exhibits high capacity during the initial charge; however, it suffers notoriously low Coulombic efficiency due to oxygen and surface activities. Here, we successfully optimize the oxygen activities toward reversible oxygen redox reactions by intentionally introducing protons into lithium octahedral vacancies in the Li2MnO3 system with its original structural integrity maintained. Combining structural probes, theoretical calculations, and resonant inelastic X-ray scattering results, a moderate coupling between the introduced protons and lattice oxygen at the oxidized state is revealed, which stabilizes the oxygen activities during charging. Such a coupling leads to an unprecedented initial Coulombic efficiency (99.2%) with a greatly improved discharge capacity of 302 mAh g-1 in the protonated Li2MnO3 electrodes. These findings directly demonstrate an effective concept for controlling oxygen activities in Li-rich systems, which is critical for developing high-energy cathodes in batteries.

Freezing interface interacting with soft cells is a core issue in cryopreservation. However, most of existing cryopreservation approaches face challenges such as complex processing and poor controllability. Herein we report a new, nanotechnology-based method to modulate interfacial interactions between ice and cells via the premelting theory. Through the interfacial design by controllably modifying biocompatible nanoshells on the surface of cells, the effective Hamaker constant between cells and ice can be modified. The thickness of premelted films between cells and ice is further regulated, and directional migration of coated cells occurs in a thermal gradient to achieve directional thawing of coated cells. The decreased mortality of freezing coated cells suggests the premelting modulation in freezing and thawing can effectively protect cells from mechanical damage of ice formation during freezing and ice recrystallization during thawing.

The advantage of wind energy is significant to the improvement of energy structure. This paper is mainly to demonstrate a bionic design for wind turbine blades inspired by the airfoil of owl wing and the herringbone groove structure of owl feathers. The blade of A 200 W horizontal axis wind turbine is taken as prototype blade. The design of bionic airfoil blade is based on 50% and 70% cross section airfoils of owl wing, which is combined with the parameters of the prototype blade. The design of bionic coupling blade is based on bionic airfoil blade, which is coupled with the herringbone groove structure. Numerical simulation is utilized to study the aerodynamic performance of all blades. These simulations utilize an incompressible Reynolds-averaged Navier–Stokes solver and shear stress transport k–ω turbulent model at different tip speed ratios (TSRs). Results show that the power coefficient (Cp) of bionic airfoil blades is higher than that of prototype blade in the TSR of 6–9; the Cp of bionic coupling blades is higher than that of bionic airfoil blades in the TSR of 6–10. The larger curvature of leading edge of blades leads to larger flow velocity, which leads to the smaller pressure on the leeward surface. The herringbone groove structure enhances the flow attachment by generating vortices, which reduces the pressure on the leeward surface of bionic coupling blades. Compared with the prototype blade and bionic airfoil blade, the pressure difference between the windward surface and the leeward surface of the bionic coupling blade is larger.

The digitisation of music and other creative industries has followed a dramatically different trajectory in China, patterned by its distinctive historical and institutional context. In contrast to the West, where established intellectual property rights holders (e.g. record labels) retained control, despite digital disruption, China has become a laboratory for sustained, large-scale experimentation with services. Though a liberal environment initially allowed many online music services to emerge, recently enforced requirements to pay license fees provoked rapid restructuring. Through merger and acquisition, the wealthy internet platforms (BAT: Baidu, AliBaba, Tencent) now dominate the market. This detailed qualitative study of the strategies and interactions between industry players, the administration and international business reveals how BAT competed to sign up content and establish ways to provide low-fee or free music services that would be sustainable in the Chinese context. They were compelled to pursue discovery-driven innovation, based on ‘trial and error’ learning, launching and adapting services at scale in an extremely turbulent setting. Building on their differing core businesses they developed diverse online music businesses, integrated with services across adjacent platforms (film, TV, games)and in lateral markets (e-commerce, payment systems) to create cross-platform service infrastructures, through which they can capture multiple value propositions.

Contains fulltext : 195184.pdf (Publisher’s version ) (Open Access) Summary The distribution and importance of anaerobic ammonium oxidation (anammox) and nitrite-dependent anaerobic methane oxidation (n-damo) have been identified in aquatic ecosystems; their role in agricultural upland soils however has not yet been well investigated. In this study, we examined spatio-temporal distributions of anammox and n-damo bacteria in soil profiles (300 cm depth) from an agricultural upland. Monitoring nitrogen (N) conversion activity using isotope-tracing techniques over the course of one year showed denitrification (99.0% N-loss in the winter and 85.0% N-loss in the summer) predominated over anammox (1.0% N-loss in the winter and 14.4% N-loss in the summer) and n-damo (0.6% N-loss in the winter) in surface soils (0?20 cm). While below 20 cm depth, N-loss was dominated by anammox (79.4 ± 14.3% in the winter and 65.4 ± 12.5% in the summer) and n-damo was not detected. Phylogenetic analysis showed that Candidatus Brocadia anammoxidans dominated the anammox community in the surface soil and Candidatus Brocadia fulgida dominated below 20 cm depth. Dissimilatory nitrate reduction to ammonium (DNRA), another nitrite reduction process, was found to play a limited role (4.9 ± 3.5%) in the surface soil compared with denitrification; below 80 cm DNRA rates were much higher than rates of anammox and denitrification. Ammonium oxidation was the main source of above 80 cm (70.9 ± 23.3%), the key influencing factor on anammox rates, and nitrate reduction (100%) was the main source below 80 cm. Considering the anammox, n-damo and denitrification rates as a whole in the sampled soil profile, denitrification is still the main N-loss process in upland soils.