Reinforcement Learning Based Quantization Strategy Optimal Assignment Algorithm for Mixed Precision

The quantization algorithm compresses the original network by reducing the numerical bit width of the model,which improves the computation speed. Because different layers have different redundancy and sensitivity to databit width. Reducing the data bit width will result in a loss of accuracy. Therefore, it is difficult to determinethe optimal bit width for different parts of the network with guaranteed accuracy. Mixed precision quantizationcan effectively reduce the amount of computation while keeping the model accuracy basically unchanged. In thispaper, a hardware-aware mixed precision quantization strategy optimal assignment algorithm adapted to low bitwidth is proposed, and reinforcement learning is used to automatically predict the mixed precision that meets theconstraints of hardware resources. In the state-space design, the standard deviation of weights is used to measurethe distribution difference of data, the execution speed feedback of simulated neural network accelerator inferenceis used as the environment to limit the action space of the agent, and the accuracy of the quantization model afterretraining is used as the reward function to guide the agent to carry out deep reinforcement learning training. Theexperimental results show that the proposed method obtains a suitable model layer-by-layer quantization strategyunder the condition that the computational resources are satisfied, and themodel accuracy is effectively improved.The proposed method has strong intelligence and certain universality and has strong application potential in thefield of mixed precision quantization and embedded neural network model deployment.

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.

Treatment of hepatoma with liposome-encapsulated adriamycin administered into hepatic artery of rats

AIM： To observe the therapeutic effects of liposomeencapsulated adriamycin （LADM） on hepatoma in comparison with adriamycin solution （FADM） and adriarnycin plus blank liposome （ADM ＋ BL） administered into the hepatic artery of rats. METHODS： LADM was prepared by pH gradient-driven method. Normal saline, FADM （2 mg/kg）, ADM＋BL （2 mg/kg）, and LADM （2 mg/kg） were injected via the hepatic artery in rats bearing liver W256 carcinosarcoma, which were divided into four groups randomly. The therapeutic effects were evaluated in terms of survival time, tumor enlargement ratio, and tumor necrosis degree. The difference was determined with ANOVA and Dunnett test and log rank test. RESULTS： Compared to FADM or ADM ＋ BL, LADM produced a more significant tumor inhibition （tumor volume ratio： 1.243±0.523 vs 1.883±0.708, 1.847±0.661, P 〈 0.01）, and more extensive tumor necrosis. The increased life span was prolonged significantly in rats receiving LADM compared with FADM or ADM＋BL （231.48 vs 74.66, 94.70） （P 〈 0.05）. CONCLUSION： The anticancer efficacies of adriamycin on hepatoma can be strongly improved by liposomal encapsulation through hepatic arterial administration.

Investigation of the redox performance of pyrite cinder calcined at different temperature in chemical looping combustion

As an industrial solid waste,pyrite cinder exhibited excellent reactivity and cycle stability in chemical looping combustion.Prior to the experiment,oxygen carriers often experienced a high temperature calcination process to stabilize the physico-chemical properties,which presented significant influence on the redox performance of oxygen carriers.However,the effect of calcination temperature on the cyclic reaction performance of pyrite cinder has not been studied in detail.In this work,the effect of calcination temperature on the redox activity and attrition characteristic of pyrite cinder were studied in a fluidizedbed reactor using CH_(4) as fuel.A series of pyrite cinder samples were prepared by controlling the calcination temperature.The redox activity and attrition rate of the obtained pyrite cinder samples were investigated deeply.The results showed that calcination temperature displayed significant impact on the redox performance of pyrite cinder.Considering CH_(4) conversion(80%–85%)and attrition resistance,the pyrite cinder calcined at 1050℃ presented excellent redox properties.In the whole experiment process,the CO_(2) selectivity of the pyrite cinder samples were not affected by the calcination temperature and were still close to 100%.The results can provide reference for optimizing the calcination temperature of pyrite cinder during chemical looping process.

Effect of supports on the redox performance of pyrite cinder in chemical looping combustion

Chemical looping combustion(CLC)is a clean and efficient flame-free combustion technology,which combust the fuels by lattice oxygen from a solid oxygen carrier with inherent CO_(2)capture.The development of oxygen carriers with low cost and high redox performance is crucial to the whole efficiency of CLC process.As the solid by-product from the sulfuric acid production,pyrite cinder presented excellent redox performance as an oxygen carrier in CLC process.The main components in pyrite cinder are Fe_(2)O_(3),CaSO_(4),Al_(2)O_(3)and SiO_(2)in which Fe_(2)O_(3)is the active component to provide lattice oxygen.In order to systematic investigate the functions of supports(CaSO_4,Al_(2)O_(3)and SiO_(2))in pyrite cinder,three oxygen carriers(Fe_(2)O_(3)-CaSO_(4),Fe_(2)O_(3)-Al_(2)O_(3)and Fe_(2)O_(3)-SiO_(2))were prepared and evaluated in this study.The results showed that Fe_(2)O_(3)-CaSO_(4) displayed high redox activity and cycling stability in the multiple redox cycles.However,both Fe_(2)O_(3)-Al_(2)O_(3)and Fe_(2)O_(3)-SiO_(2)experienced serious deactivation during redox reactions.It indicated that the inert Fe-Si solid solution(Fe_(2)SiO_(4))was formed in the spent Fe_(2)O_(3)-SiO_(2)sample,which decreased the oxygen carrying capacity of this sample.The XPS results showed that the oxygen species on the surface of Fe_(2)O_(3)-CaSO_(4) could be fully recovered after the 20 redox cycles.It can be concluded that CaSO_(4) is the key to the high redox activity and cycling stability of pyrite cinder.

企业会使用慈善捐赠响应强制导向企业社会信息制度吗？——基于准实验研究的证据

为了督促企业社会责任活动,越来越多的经济体推行强制导向的企业社会责任信息披露制度,2009年我国也加入此列。本文围绕此次信息制度导向变更事件,以2006至2008年作为前政策窗口期,2009至2011年作为政策效应观察窗口期,基于沪深两市上市企业慈善捐赠数据开展准实验研究。研究结果表明,在强制披露制度施行后,强制披露企业对慈善捐赠的投入水平更高。

Skin-inspired electronics:emerging semiconductor devices and systems

Current electronics are driven by advanced microfabrication for fast and efficient information processing.In spite of high performance,these wafer-based devices are rigid,non-degradable,and unable to autonomous repair.Skin-inspired electronics have emerged as a new class of devices and systems for next-generation flexible and wearable electronics.The technology gains inspiration from the structures,properties,and sensing mechanisms of the skin,which may find a broad range of applications in cutting-edge fields such as healthcare monitoring,human-machine interface,and soft robotics/prostheses.Practical demands have fueled the development of electronic materials with skin-like properties in terms of stretchability,self-healing capability,and biodegradability.These materials provide the basis for functional sensors with innovative and biomimetic designs.Further system-level integrations and optimizations enable new forms of electronics for real-world applications.This review summarizes recent advancements in this active area and speculates on future directions.

商业银行会计业务操作风险分析

商业银行会计业务的开展,能够全面精确地核算并监督商业银行的运营管理活动,与商业银行账务的准确度存在相关关系。商业银行会计核算工作的开展,需要进行严格把控,防范各环节纰漏,以降低商业银行会计风险。会计业务监督不到位,导致金融系统案件频发。文章就商业银行会计业务操作风险控制的重要性进行阐述,明确商业银行会计业务风险存在的表现形式,分析商业银行会计业务操作风险成因,进一步探究商业银行会计业务操作风险的防范对策,旨在促进商业银行的稳定健康发展。

Healthcare electronic skin devices

Electronic skin(e-skin),a kind of flexible sensor arrays and system that mimic the properties and sensing functions of human skin,represents a new paradigm of sensing and control(Fig.1).The noun of'skin electronics'made its debut in the Sensitive Skin Workshop organized jointly by the National Science Foundation and Defense Advanced Research Projects Agency of USA in October 1999 in Arlington.

Research Progress on Photocatalytic/Photoelectrocatalytic Oxidation of Nitrogen Oxides

The emission of nitrogen oxides(NO_(x))increases year by year,causing serious problems to our livelihoods.The photocatalytic oxidation of NO_(x)has attracted more attention recently because of its effi cient removal of NO_(x),especially for low concentra-tions of NO_(x).In this review,the mechanism of the photocatalytic oxidation of NO_(x)is described.Then,the recent progress on the development of photocatalysts is reviewed according to the categories of inorganic semiconductors,bismuth-based compounds,nitrogen carbide polymer,and metal organic frameworks(MOFs).In addition,the photoelectrocatalytic oxida-tion of NO_(x),a method involving the application of an external voltage on the photocatalytic system to further increase the removal effi ciency of NO_(x),and its progress are summarized.Finally,we outline the remaining challenges and provide our perspectives on the future directions for the photocatalytic oxidation of NO_(x).

安顺市野生足茎毛兰资源调查及受胁原因分析

为了解种群现状和保护种群扩繁,对安顺境内野生足茎毛兰(Eria coronaria)资源进行调查。结果表明:(1)安顺境内野生足茎毛兰花期为9月下旬至11月下旬,与原文献记载不符;资源数量约8500株,分布面积为1400m^2;(2)主要分布在海拔1000m以上的喀斯特石山中、上部,植被郁闭度0. 6以上,林下覆盖度在40%～60%间,且坡度在35°以上的陡坡石面或石沟中;(3)足茎毛兰受胁原因分为外因和内因,其中内因主要是种群结实率、种子繁殖力等;外因主要表现在对生境要求、人为活动、传粉媒介专一性等方面。

数字财务推动企业战略落地——海通证券数字财务创新实践案例

随着数字经济的到来,大数据、人工智能、移动互联网、云计算、物联网等技术成为企业发展的新动能,也催生了财务管理的新模式。数字化技术能够让财务更好地发挥企业天然数据中心的优势,从后台走到前台,为企业提供以数据驱动为核心的预测与决策支持、业务支持以及风险控制,支撑企业战略执行和高质量发展。作为国内领先的证券公司之一,海通证券较早开始企业数字化建设,而财务数字化转型是海通证券股份有限公司(以下简称“海通证券”)数字化战略的切入点和着力点。近年来,海通证券紧紧围绕国家“十四五”规划和公司的整体发展战略,运用前沿科技,积累丰富数据,挖掘应用场景,开展系统整合,逐步构建数字财务体系,努力从提升管理能级、助力业务发展、强化风险控制等方面发挥数字财务的威力,全方位多角度地支撑企业“集团化、国际化、信息化”的战略,为证券行业的数字财务实践探索提供了有价值的参考和借鉴。

财务数字化赋能管理会计实践

2022年2月的一个午后,《中国管理会计》期刊走进海通证券大厦,专程与海通证券股份有限公司(以下简称“海通证券”)计划财务部总经理马中进行了一场深度对话,规划和设计的对话主题为:财务数字化赋能管理会计实践。潘飞:马总下午好!随着政府主导管理会计体系化建设的发展,2017年7月《中国管理会计》杂志正式创刊,在此首先转达中国总会计师协会刘红薇会长的问候,并对海通证券和马总对我们期刊给予的大力支持表示由衷的感谢。

财务数字化赋能管理会计研究—海通证券数字财务实践

以“大、智、移、云、物、区”为代表的新一代财务数字化技术正在深度影响着管理会计的走向。基于管理会计展开的企业流程重组、决策支持、组织架构重塑已成为企业财务数字化转型的重要组成部分。近年来,海通证券紧紧围绕国家“十四五”发展规划和公司集团化、国际化、信息化的发展战略,以“风险管控+业财融合+精细化管理”为目标,以“科技+数据+场景”为驱动,依靠内生动力进行系统整合,通过全面深化财务数字化技术赋能管理会计,切实推进业财融合,防范财务风险,支持经营决策,服务战略发展,促进财务转型和职能转变,全面提升管理能级。

Effect of doping order on metal-free heteroatoms dual-doped carbon as oxygen reduction electrocatalyst

Metal-free heteroatoms dual-doped carbon has been recognized as one of the most promising Pt/C-substitutes for oxygen reduction reaction(ORR).Herein,we optimize the preparation process by doping order of metal-free heteroatoms to obtain the best electrocatalytic performance through three types of dual-doped carbon,including XC-N(first X doping then N doping),NC-X(first N doping then X doping) and NXC(N and X doping)(X=P,S and F).XC-N has more defect than the other two indicated by Raman spectra.X-ray photoelectron spectrom(XPS) measurements indicate that N and X have been dual-doped into the carbon matrix with different doping contents and modes,Electrocatalytic results,including the potential of ORR peak(Ep),the half-wave potential,the diffusion-limiting current density mainly follows the order of XC-N>NC-X> NXC,Furthermore,the synergistic effect of second atom doping are also compared with the single doped carbon(NC,PC,SC and FC).The differences in electronegativity and atomic radius of these metal-free heteroatoms can affect the defect degree,the doping content and mode of hete roatoms on carbon matrix,induce polarization effect and space effect to affect O2 adsorption and product desorption,ultimately to the ORR electrocatalytic performance.

Elevated CO_2 Accelerates Phosphorus Depletion by Common Bean(Phaseolus vulgaris) in Association with Altered Leaf Biochemical Properties

Phosphorus （P） is a major limiting factor for plant productivity in many ecosystems and agriculture. The projected increase in atmospheric CO2 is likely to result in changes in plant mineral consumption and growth. We studied P depletion by common bean （Phaseolus vulgaris） cultured hydroponically under ambient （377±77μmol mol^-1） or elevated （650±32 μmol mol^-1） CO2 in media of low or high P. Under elevated CO2 compared to ambient CO2, the maximum P depletion rate increased by 98% at low P and 250% at high P, and P was depleted about 2-5 weeks sooner; leaf acid phosphatase （APase） activity and chlorophyll content both increased significantly; root-to-shoot ratio increased significantly at high P, although it was unaffected at low P; lateral root respiration rate showed no change, suggesting that COs did not affect P depletion via metabolic changes to the roots; the total biomass at final harvest was significantly higher at both low and high P. Our data showed that the increased rate and amount of P depletion during plant growth under elevated CO2 occurred in association with alterations in leaf biochemical properties, i.e., enhanced activities of leaf APase and increased leaf chlorophyll content.