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。     

Fully asynchronous stochastic coordinate descent: a tight lower bound on the parallelism achieving linear speedup

Mathematical Programming - We seek tight bounds on the viable parallelism in asynchronous implementations of coordinate descent that achieves linear speedup. We focus on asynchronous coordinate...     

Close‐Packed Two‐Dimensional Silver Nanoparticle Arrays: Quadrupolar and Dipolar Surface Plasmon Resonance Coupling

Silver nanoparticles (NPs) ranging in size from 40 to 100 nm were prepared in high yield by using an improved seed‐mediated method. The homogeneous Ag NPs were used as building blocks for 2D assembled Ag NP arrays by using an oil/water interface. A close‐packed 2D array of Ag NPs was fabricated by using packing molecules (3‐mercaptopropyltrimethoxysilane) to control the interparticle spacing. The homogeneous 2D Ag NP array exhibited a strong quadrupolar cooperative plasmon mode resonance and a dipolar red‐shift relative to individual Ag NPs suspended in solution. A well‐arranged 2D Ag NP array was embedded in polydimethylsiloxane film and, with biaxial stretching to control the interparticle distance, concomitant variations of the quadrupolar and dipolar couplings were observed. As the interparticle distance increased, the intensity of the quadrupolar cooperative plasmon mode resonance decreased and dipolar coupling completely disappeared. The local electric field of the 2D Ag NP array was calculated by using finite difference time domain simulation and qualitatively showed agreement with the experimental measurements.     

无序激光晶体及其超快激光研究进展

激光是受激辐射的光放大,所辐射的波长取决于增益介质中关键电子的能级结构,特别是其最外层电子的状态决定了可能实现的激光特性。激光发展60年来,激光晶体作为激光的重要激活材料,推动了激光技术的进步和普及,是一个研究历史长而又异常活跃的研究领域。当前,超短超强脉冲激光在加工、医疗、国防等关系国计民生的领域有重要需求,适合超短超强激光的激光晶体成为了本领域的研究热点,其关键是揭示最外层电子的影响因素及设计和生长具有宽波段发射性能的激光晶体。本论文从探讨影响激活离子光谱性能的关键因素出发,综述了以本课题组十余年研究的10余种无序激光晶体为主要部分的研究结果和进展,涉及晶体生长、晶体物理、激光器件设计及应用等工作,包含了高级、中级和低级对称性晶体及其获得的最短脉冲激光结果,希望能为本领域的后续研究提供一定的参考和借鉴。     

新型4-羟色胺类化合物的合成及生物活性

以4-羟基吲哚为原料,经吲哚环4位乙酰基化、3位亲电取代、酰胺化和还原加氢等反应合成目标化合物7.通过核磁共振氢谱及碳谱对化合物进行结构表征,并对目标化合物进行体外抗氧化生物活性测试.初步生物活性测试结果表明,化合物7a,7b,7c和7d对DPPH·均有很强的清除作用(清除率为85.25%~90.73%),7e,7f,7g,7h作用较差;目标化合物与Vc相比,对·OH的清除作用稍差,最高清除率25.66%(Vc的最高清除率为34.67),但各化合物整体水平相当;在清除O-2·能力上化合物7a,7d,7g,7h最大清除率(分别为19.34,35.35,27.93和31.74)均强于同等浓度的Vc(17.58).     

Asymptotic Behavior of Least Energy Solutions for a Fractional Laplacian Eigenvalue Problem on ℝN

Acta Mathematica Sinica, English Series - We are interested in the existence and asymptotic behavior for the least energy solutions of the following fractional eigenvalue problem $$\left({\rm{P}}...