基于快速神经网络架构搜索的鲁棒图像水印网络算法

为解决深度学习在图像水印算法中计算量大且模型冗余的问题，提高图像水印算法在抵抗噪声、旋转和剪裁等攻击时的鲁棒性，提出基于快速神经网络架构搜索（neural architecture search，NAS）的鲁棒图像水印网络算法。通过多项式分布学习快速神经网络架构搜索算法，在预设的搜索空间中搜索最优网络结构，进行图像水印的高效嵌入与鲁棒提取。首先，将子网络中线性连接的全卷积层设置为独立的神经单元结构，并参数化表示结构单元内节点的连接，预先设定结构单元内每个神经元操作的搜索空间；其次，在完成一个批次的数据集训练后，依据神经元操作中的被采样次数和平均损失函数值动态更新概率；最后，重新训练搜索完成的网络。水印网络模型的参数量较原始网络模型缩减了92%以上，大大缩短了模型训练时间。由于搜索得到的网络结构更为紧凑，本文算法具有较高的时间性能和较好的实验效果，在隐藏图像时，对空域信息的依赖比原始网络更少。对改进前后的2个网络进行了大量鲁棒性实验，对比发现，本文算法在CIFAR-10数据集上对抵抗椒盐噪声和旋转、移除像素行（列）等攻击优势显著；在ImageNet数据集上对抵抗椒盐高斯噪声、旋转、中值滤波、高斯滤波、JPEG压缩、裁剪等攻击优势显著，特别是对随机移除行（列）和椒盐噪声有较强的鲁棒性。     

Flow-induced shear stress and deformation of a core–shell-structured microcapsule in a microchannel

A numerical model was developed and validated to investigate the fluid–structure interactions between fully developed pipe flow and core–shell-structured microcapsule in a microchannel. Different flow rates and microcapsule shell thicknesses were considered. A sixth-order rotational symmetric distribution of von Mises stress over the microcapsule shell can be observed on the microcapsule with a thinner shell configuration, especially at higher flow rate conditions. It is also observed that when being carried along in a fully developed pipe flow, the microcapsule with a thinner shell tends to accumulate stress at a higher rate compared to that with a thicker shell. In general, for the same microcapsule configuration, higher flow velocity would induce a higher stress level over the microcapsule shell. The deformation gradient was used to capture the microcapsule's deformation in the present study. The effect of Young's modulus on the microcapsule shell on the microcapsule deformation was investigated as well. Our findings will shed light on the understanding of the stability of core–shell-structured microcapsule when subjected to flow-induced shear stress in a microfluidic system, enabling a more exquisite control over the breakup dynamics of drug-loaded microcapsule for biomedical applications.     

Logarithmic Regge pole

This work presents the subtraction procedure and the Regge cut in the logarithmic Regge pole approach. The subtraction mechanism leads to the same asymptotic behavior as previously obtained in the non-subtraction case. The Regge cut, in contrast, introduces a clear role to the non-leading contributions for the asymptotic behavior of the total cross-section. From these results, some simple parameterization is introduced to fit the experimental data for the proton-proton and antiproton-proton total cross-section above some minimum value up to the cosmic-ray. The fit parameters obtained are used to present predictions for the \begin{document}$ \rho(s)$\end{document} -parameter as well as to the elastic slope \begin{document}$ B(s)$\end{document} at high energies.     

Role of FTFSI Anion Asymmetry on Physical Properties of AFTFSI (A=Li,Na and K) Based Electrolytes and Consequences on Supercapacitor Application

This study compares the physicochemical properties of six electrolytes comprising of three salts: LiFTFSI, NaFTFSI and KFTFSI in two solvent mixtures, the binary (3EC/7EMC) and the ternary (EC/PC/3DMC). The transport properties (conductivity, viscosity) as a function of temperature and concentration were modeled using the extended Jones-Dole-Kaminsky equation, the Arrhenius model, and the Eyring theory of transition state for activated complexes. Results are discussed in terms of ionicity, solvation shell, and cross-interactions between electrolyte components. The application of the six formulated electrolytes in symmetrical activated carbon (AC)//AC supercapacitors (SCs) was characterized by cyclic voltammetry (CV), galvanostatic cycling with potential limitation (GCPL), electrochemical impedance spectroscopy (EIS) and accelerated aging. Results revealed that the geometrical flexibility of the FTFSI anion allows it to access and diffuse easily in AC whereas its counter ions (Li⁺, Na⁺ or K⁺) can remain trapped in porosity. However, this drawback was partially resolved by mixing LiFTFSI and KFTFSI salts in the electrolyte.     

Theoretical study on the mechanism for the excited-state double proton transfer process of an asymmetric Schiff base ligand

Excited-state double proton transfer (ESDPT) in the 1-[(2-hydroxy-3-methoxy-benzylidene)-hydrazonomethyl]-naphthalen-2-ol (HYDRAVH₂) ligand was studied by the density functional theory and time-dependent density functional theory method. The analysis of frontier molecular orbitals, infrared spectra, and non-covalent interactions have cross-validated that the asymmetric structure has an influence on the proton transfer, which makes the proton transfer ability of the two hydrogen protons different. The potential energy surfaces in both S₀ and S₁ states were scanned with varying O-H bond lengths. The results of potential energy surface analysis adequately proved that the HYDRAVH₂ can undergo the ESDPT process in the S₁ state and the double proton transfer process is a stepwise proton transfer mechanism. Our work can pave the way towards the design and synthesis of new molecules.     

Experimental and numerical analysis of cylindrical turbulators in a double pipe heat exchanger under turbulent flow conditions

In the present work, the use of cylindrical turbulators in a double pipe heat exchanger has been investigated. Cylindrical fin type of turbulators has been placed circumferentially separated by 90° on the outer side of an inner pipe at a regular pitch. Experimental studies were undertaken for different air flow rates in a turbulent regime whose Reynolds number range between 2500 and 10000. Heat transfer characteristics like Nu and friction factor have been experimentally determined. Parametric studies were conducted by changing the pitch and also the orientation of the turbulators. Nu and friction factor were found to increase as the pitch is reduced. A model with alternatively changed orientation outperformed others by exhibiting highest Nu and reduced friction factor.     

ON THE CONVERGENCE OF MULTIPLICATIVE ITERATIVE ALGORITHMS WITH INEXACT LINE SEARCH

1.IntroductionIntillspaperweanalyzetheconvergenceonmultiplicativeiterativealgorithmsfortheIninimizationofadiffcrentiablefunctiondefinedonthepositiveorthantofR".ThealgorithmissllggestedbyEggermolltl'],andisrelatedtotheEM[2](Expextation--Maximization)algoritllnlforPositronemissiontonlography[']andimagereconstructi..14].Wecollsidertheproblenl"linf(x)s.t.x20.Themultiplicativeiterativealgorithmshavethel'orlniforj=l,2,',n,withAhdeterminedthroughalinesearch.Whilelusem[5]establishedanelegantconv…     

Local and Parallel Finite Element Algorithms for Eigenvalue Problems

Abstract Some new local and parallel finite element algorithms are proposed and analyzed in this paper foreigenvalue problems.With these algorithms, the solution of an eigenvalue problem on a fine grid is reduced tothe solution of an eigenvalue problem on a relatively coarse grid together with solutions of some linear algebraicsystems on fine grid by using some local and parallel procedure.A theoretical tool for analyzing these algorithmsis some local error estimate that is also obtained in this paper for finite element approximations of eigenvectorson general shape-regular grids.     

一种基于GIS的选址问题及算法

本文考虑了选址区域内存在地理阻断情况下的一种基于GIS的选址问题.对单配送中心选址模型,以GIS返回的任意两点间的最短可行路径的长度作为修正距离函数,分析了目标函数在凸形选址区域上非凸非连续的性质.进一步,采用给出了一种近似搜索算法并通过一个实例计算与重心法进行了比较.