A Linear Frequency Principle Model to Understand the Absence of Overfitting in Neural Networks

Why heavily parameterized neural networks(NNs) do not overfit the data is an important long standing open question. We propose a phenomenological model of the NN training to explain this non-overfitting puzzle. Our linear frequency principle(LFP) model accounts for a key dynamical feature of NNs: they learn low frequencies first, irrespective of microscopic details. Theory based on our LFP model shows that low frequency dominance of target functions is the key condition for the non-overfitting of NNs and is verified by experiments. Furthermore,through an ideal two-layer NN, we unravel how detailed microscopic NN training dynamics statistically gives rise to an LFP model with quantitative prediction power.     

高温吸热管内超临界CO2传热特性的数值模拟

研究超临界CO2在高温吸热管内的传热特性是将其应用于聚光太阳能热发电技术中的基础.本文对此进行了数值模拟研究,分析了流体温度、流动方向、系统压力、质量流率和热流密度对对流传热系数和Nu数的影响.结果表明:高温区(800—1050 K)的对流传热系数和Nu数受流动方向和系统压力的影响均很小,但都随着质量流率的增大以及热流...     

微腔内气体抽离的多尺度模拟与分析

基于适用于整个克努森数范围的流动理论,建立了去除惯性约束聚变实验中靶丸内空气的理论模型,并设计实验验证了此模型的可靠性。物理实验要求靶丸内空气浓度低于10×10⁻⁶,数值模拟了去除靶丸内空气的过程,重点分析了靶丸内空气浓度、压力与除气时间的关系。计算并比较了单管路一次抽气法、单管路循环抽气法与双管路流洗法三种去除靶丸内空气方法的时间成本。数值计算结果表明：单管路一次抽气法中,靶丸上的微通道的存在对去除靶丸内空气所需时间的影响不可忽略,在考虑靶丸上微通道与充气管的情况下,需要1961.77 h才能使靶丸内的空气浓度达到标准。单管路循环抽气法中,抽气次数与单次抽气程度会影响去除靶丸内空气所需总时间,在单次抽气程度值取最优的情况下,采用充三次,抽四次的方案可使达标总时间减少至1 h左右,此方案下单次充气和抽气时间分别为6 min和10 min。而采用双管路流洗法则仅需11 min便可使靶丸内空气浓度达标。     

电子照相技术成像模拟研究

针对高能电子照相的关键物理过程开展蒙特卡罗照相模拟研究,采用Geant4围绕2.5 GeV电子在四极透镜组内的输运、电子与物质相互作用衰减等照相基本过程展开研究。通过设计不同材料、不同厚度的含缺陷平板作为模拟照相客体,开展放大型电子照相系统缺陷分辨能力模拟。此外,采用不同材料、不同厚度的台阶样品,模拟获得了电子束流穿过相应面密度材料后的线扩展函数,进一步评估电子照相对实心客体的探测分辨能力。     

平台间光束耦合传输与控制技术研究

介绍了平台间光路耦合传输系统的构成及光轴稳定控制的实现方法,开展了耦合校正系统和探测控制系统的设计,对校正系统进行了动态范围和模态仿真,优化设计后研制出光束耦合传输与控制系统。在对快反镜性能参数测试之后,开展了平台间光束耦合传输与控制实验,当振动台加载0 db振动谱且控制系统开环时,光轴X轴抖动10.9″@RSM,Y轴抖动102.3″@RSM,闭环时,光轴X轴抖动0.75″@RSM,Y轴抖动1.11″@RSM,通过频谱分析发现,快反镜光轴耦合系统闭环时对28 Hz以内光轴抖动具有较好地抑制作用,在系统开环残差较大的频率段2～6 Hz的抑制比为−40～−30 dB。实验结果表明,该光轴耦合控制系统对平台间光束传输过程中光束抖动具有较好地抑制和稳定效果。     

Improved efficiency and stability of perovskite solar cells with molecular ameliorating of ZnO nanorod/perovskite interface and Mg-doping ZnO

Despite the advanced efficiency of perovskite solar cells(PSCs),electron transportation is still a pending issue.Here the polymer polyvinylpyrrolidone(PVP)is used to enhance the electron injection,which is thanks to the passivation of the defects at the interface between the ZnO electron transporting layer(ETL)and the perovskite.The application of the PVP layer inhibits the device degradation,and 80%of the primary efficiency is kept after 30 d storage in air condition.Additionally,the efficiency of the device is further enhanced by improving the conductivity and crystallinity of the ZnO ETL via Magnesium(Mg)doping in the ZnO nanorods(ZnO NRs).Moreover,the preparation parameters of the ZnO NRs are optimized.By employing the high-crystallinity ZnO ETL and the PVP layer,the power conversion efficiency(PCE)of the champion device is increased from 16.29%to 19.63%.These results demonstrate the advantages of combining mesoscale manipulation with interface modification and doping together.     

Chaotic Dynamics of a Josephson Junction with a Ratchet Potential and Current-Modulating Damping

The chaotic dynamics of a Josephson junction with a ratchet potential and current-modulating damping are studied. Under the first-order approximation, we construct the general solution of the first-order equation whose boundedness condition contains the famous Melnikov chaotic criterion. Based on the general solution, the incomputability and unpredictability of the system’s chaotic behavior are discussed. For the case beyond perturbation conditions, the evolution of stroboscopic Poincaré sections shows that the system undergoes a quasi-periodic transition to chaos with an increasing intensity of the rf-current. Through a suitable feedback controlling strategy, the chaos can be effectively suppressed and the intensity of the controller can vary in a large range. It is also found that the current between the two separated superconductors increases monotonously in some specific parameter spaces.     

One-step preparation of branched PEG functionalized AIE-active luminescent polymeric nanoprobes

The synthesis of amphiphilic aggregation-induced emission(AIE) dyes based organic nanoparticles has recently attracted increasing attention in the biomedical fields. These AIE dyes based nanoparticles could effectively overcome the aggregation caused quenching effect of conventional organic dyes, making them promising candidates for fabrication of ultrabright organic luminescent nanomaterials. In this work, AIE-active luminescent polymeric nanoparticles(4-NH_2-PEG-TPE-E LPNs) were facilely fabricated through Michael addition reaction between tetraphenylethene acrylate(TPE-E) and 4-arm-poly(ethylene glycol)-amine(4-NH_2-PEG) in rather mild ambient. The 4-NH_2-PEG can not only endow these AIE-active LPNs good water dispersibility, but also provide functional groups for further conjugation reaction. The size, morphology and luminescent properties of 4-NH_2-PEG-TPE-E LPNs were characterized by a series of techniques in detail. Results suggested that these AIE-active LPNs showed spherical morphology with diameter about 100–200 nm. The obtained 4-NH_2-PEG-TPE-E LPNs display high water dispersibility and strong fluorescence intensity because of their self assembly and AIE properties of TPE-E.Biological evaluation results demonstrated that 4-NH_2-PEG-TPE-E LPNs showed negative toxicity toward cancer cells and good fluorescent imaging performance. All of these features make 4-NH_2-PEG-TPE-E LPNs promising candidates for biological imaging and therapeutic applications.     

Reverse depletion effects and the determination of ligand density on some spherical bioparticles

In cell environments crowded with macromolecules, the depletion effects act and assist in the assembly of a wide range of cellular structures, from the cytoskeleton to the chromatin loop, which are well accepted. But a recent quantum dot experiment indicated that the dimensions of the receptor–ligand complex have strong effects on the size-dependent exclusion of proteins in cell environments. In this article, a continuum elastic model is constructed to resolve the competition between the dimension of the receptor–ligand complex and depletion effects in the endocytosis of a spherical virus-like bioparticle. Our results show that the depletion effects do not always assist endocytosis of a spherical virus-like bioparticle; while the dimension of the ligand–receptor complex is larger than the size of a small bioparticle in cell environments, the depletion effects do not work and reverse effects appear. The ligand density covered on the virus can be identified quantitatively.