Activation Redistribution Based Hybrid Asymmetric Quantization Method of Neural Networks

The demand for adopting neural networks in resource-constrained embedded devices is continuously increasing.Quantization is one of the most promising solutions to reduce computational cost and memory storage on embedded devices.In order to reduce the complexity and overhead of deploying neural networks on Integeronly hardware,most current quantization methods use a symmetric quantization mapping strategy to quantize a floating-point neural network into an integer network.However,although symmetric quantization has the advantage of easier implementation,it is sub-optimal for cases where the range could be skewed and not symmetric.This often comes at the cost of lower accuracy.This paper proposed an activation redistribution-based hybrid asymmetric quantizationmethod for neural networks.The proposedmethod takes data distribution into consideration and can resolve the contradiction between the quantization accuracy and the ease of implementation,balance the trade-off between clipping range and quantization resolution,and thus improve the accuracy of the quantized neural network.The experimental results indicate that the accuracy of the proposed method is 2.02%and 5.52%higher than the traditional symmetric quantization method for classification and detection tasks,respectively.The proposed method paves the way for computationally intensive neural network models to be deployed on devices with limited computing resources.Codes will be available on https://github.com/ycjcy/Hybrid-Asymmetric-Quantization.

Asymmetric orbital hybridization in Zn-doped antiperovskite Cu_(1-x)Zn_(x)NMn_(3)enables highly efficient electrocatalytic hydrogen production

Rational design of efficient and robust earth-abundant alkaline hydrogen evolution reaction(HER)catalysts is a key factor for developing energy conversion technologies.Currently,antiperovskite nitride CuNMn_(3)has garnered significant interest due to its remarkable properties such as negative/zero thermal expansion and magnetocaloric effects.However,when utilized as hydrogen evolution catalysts,it encounters large challenge resulting from excessively strong/weak interactions with adsorbed H on Mn/Cu active sites,which leads to low HER activity.In this study,we introduce an asymmetric orbital hybridization strategy in Zn-doped Cu_(1-x)Zn_(x)NMn_(3)by leveraging the localization of Zn electronic states to reconfigure the electronic structures of Cu and Mn,thereby reducing the energy barrier for water dissociation and optimizing Cu and Mn active sites for hydrogen adsorption and H_(2)production.Electrochemical evaluations reveal that Cu_(0.85)Zn_(0.15)NMn_(3)with x=0.15 demonstrates exceptional electrocatalytic activity in alkaline electrolytes.A low overpotential of 52 mV at 10 mA cm^(-2)and outstanding stability over a 150-h test period are achieved,surpassing commercial Pt/C.This research offers a novel strategy for enhancing HER performance by modulating asymmetric hybridization of electron orbitals between multiple metal atoms within a material structure.

3D hierarchical oxygen-deficient AlCoNi-(oxy)hydroxides/N-doped carbon hybrids enable efficient battery-type asymmetric supercapacitor

Polynary transition-metal layered hydroxides are promising energy materials owing to their unique architecture,impressive theoretical capacities,and adjustable compositions.Regulating the dimensional morphology and active sites/redox states are the keys to electrochemical performance enhancement.Distinguish from the reported mono-metal or binary-metal configurations,a new ternary-metal AlCoNi-LTH is coanchored onto a highly graphitized porous N-doped carbon matrix to develop superior 3D hierarchical microporous functional energy hybrids AlCoNi-LTHs/NAC.The constructed hybrids possess superior structural durability,good electrical conductivity,and rich active sites due to the strong interfacial conjunction and favorable synergistic effect between the doped porous carbon and AlCoNi nanosheets.Consequently,the AlCoNi-LTHs/NAC hybrids demonstrate high conductivity,reasonable specific surface area,and superior specific capacitance,and the assembled hybrid battery-type supercapacitor reveals an ideal energy density of 72.6 Wh kg^(-1)at a power density of 625 W kg^(-1),which is superior to the reported devices.This strategy opens a platform to rationally design polynary transition-metal layered hydroxides and their hybrids for efficient supercapacitors.

Progress of chromosome engineering mediated by asymmetric somatic hybridization

Plant somatic hybridization has progressed steadily over the past 35 years. Many hybrid plants have been generated from fusion combinations of different phylogenetic species, some of which have been utilized in crop breeding programs. Among them, asymmetric hybrid, which usually contains a fraction of alien genome, has received more attention because of its importance in crop improvement. However, few studies have dealt with the heredity of the genome of somatic hybrid for a long time, which has limited the progress of this approach. Over recent ten years, along with the development of an effective cytogenetical tool ＂in situ hybridization （ISH）＂, asymmetric fusion of common wheat （Triticum aestivum L.） with different grasses or cereals has been greatly developed. Genetics, genomes, functional genes and agricultt, ral traits of wheat asymmetric hybrids have been subject to systematic investigations using gene cloning, genomic in situ hybridization （GISH） and molecular makers. The future goal is to fully elucidate the functional relationships among improved agronomic traits, the genes and underlying molecular mechanisms, and the genome dynamics of somatic introgression lines. This will accelerate the development of elite germplasms via somatic hybridization and the application of these materials in the molecular improvement of crop plants.

Asymmetric reinforcement in Lucania killifish:assessing reproductive isolation when both sexes choose

Reinforcement can occur when maladaptive hybridization in sympatry favors the evolution of con-specific preferences and target traits that promote behav ioral isolation(BI).In many systems,enhanced BI is due to increased female preference for conspecifics.In others,BI is driven by male preference,and in other systems both sexes exert preferences.Some of these patterns can be atributed to classic sex-specific costs and benefits of preference.Alternatively,sex differences in conspecific preference can emerge due to asymmetric postzygotic isolation(e.g.,hybrid offspring from female A × male B have lower fitness than hybrid offspring from female B × male A),which can lead to asymmetric BI(e.g.,female A and male B are less likely to mate than female B and male A).Understanding reinforcement requires understanding how conspecific preferences evolve in sympatry.Yet,estimating conspecific preferences can be difficult when both sexes are choosy.In this study,we use Lucania killifish to test the hypothesis that patterns of reinforcement are driven by asymmetric postzygotic isolation between species.If true,we predicted that sympatric female Lucania goodei and sympatric male L.parva should have lower levels of BI compared with their sympatric counterparts,as they produce hybrid offspring with the highest fitness.To address the problem of measuring BI when both sexes are choosy,we inferred the contribution to BI of each partner using assays where one sex in the mating pair comes from an allopatric population with potentially low preference,whereas the other comes from a sympatric population with high prefer-ence.For one hybrid cross direction,we found that both female L.parva and male L.goodei have high contributions to BI in sympatry.In the other hybrid cross direction,we found that only female L.goodei contribute to BI.Sympatric male L.parva readily engaged in hybrid spawnings with allo-patric L.goodei females.These results indicate that both asymmetric postzygotic isolation and the traditional sex-specific costs to preference likely affect the nature of selection on conspecific prefer-ences and target traits.