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。     

Hyperparameter on-line learning of stochastic resonance based threshold networks

Aiming at training the feed-forward threshold neural network consisting of nondifferentiable activation functions, the approach of noise injection forms a stochastic resonance based threshold network that can be optimized by various gradient-based optimizers. The introduction of injected noise extends the noise level into the parameter space of the designed threshold network, but leads to a highly non-convex optimization landscape of the loss function. Thus, the hyperparameter on-line learning procedure with respective to network weights and noise levels becomes of challenge. It is shown that the Adam optimizer, as an adaptive variant of stochastic gradient descent, manifests its superior learning ability in training the stochastic resonance based threshold network effectively. Experimental results demonstrate the significant improvement of performance of the designed threshold network trained by the Adam optimizer for function approximation and image classification.     

化学类拔尖学生培养与高校青年教师成长协同发展实践*

在高等教育大众化的背景下,通过教学改革培养高素质的拔尖人才已显得非常重要。介绍一种以新入职的青年教师为实施主体的“理论课助教+实验课教学+科研训练”相结合的教学形式。通过连续、跟踪式的教学,不但可以选拔、培养出拔尖的化学类本科生,而且能够促进新入职的青年教师快速成长进步,起到“教学相长”的作用。     

Pattern transition and regulation in a subthalamopallidal network under electromagnetic effect

Although the significant roles of magnetic induction and electromagnetic radiation in the neural system have been widely studied, their influence on Parkinson's disease (PD) has yet to be well explored. By virtue of the magnetic flux variable, this paper studies the transition of firing patterns induced by magnetic induction and the regulation effect of external magnetic radiation on the firing activities of the subthalamopallidal network in basal ganglia. We find: (i) The network reproduces five typical waveforms corresponding to the severity of symptoms: weak cluster, episodic, continuous cluster, episodic, and continuous wave. (ii) Magnetic induction is a double-edged sword for the treatment of PD. Although the increase of magnetic coefficient may lead the physiological firing activity to transfer to pathological firing activity, it also can regulate the pathological intensity firing activity with excessive β-band power transferring to the physiological firing pattern with weak β-band power. (iii) External magnetic radiation could inhibit continuous tremulous firing and β-band power of subthalamic nucleus (STN), which means the severity of symptoms weakened. Especially, the bi-parameter plane of the regulation region shows that a short pulse period of magnetic radiation and a medium level of pulse percentage can well regulate pathological oscillation. This work helps to understand the firing activity of the subthalamopallidal network under electromagnetic effect. It may also provide insights into the mechanisms behind the electromagnetic therapy of PD-related firing activity.     

Na原子链表面等离激元特性的含时密度泛函研究

本文用含时密度泛函理论研究了线性Na原子链的表面等离激元机理.主要在原子尺度下模拟计算了体系随着原子数增加及原子间距变化的集体激发过程.研究发现线性原子链有一个普遍的特性——存在一个纵模和两个横模.两个横模一般在实验上很难被观测到.纵模随着原子链长度增加,能量红移的同时,该纵模主峰的强度呈线性增长.随着原子个数的增加,端点模式(TE)开始蓝移,能量和偶极强度都逐渐趋向饱和.横模能量被劈裂的原因概括如下:(一)每个位置的电子受到的势不同,在两端的电子受到的势要比在中间的电子受到的势要高,因此两端的电荷积累也比中间多;(二)端点存在悬挂键,所以中间的电子-电子间相互作用与端点的不一样,这两方面又都与原子间距d有关.     

基于CEEMD的重大事件对香港住宅价格影响的实证分析

应用互补集成经验模态分解（CEEMD）方法对香港1997―2018年的住宅价格月度数据进行了分解，将经过重构后的数据分成高频序列、低频序列与残差项。将BP多断点检测应用于低频序列，并结合样本时段内的重大事件进行实证分析。结果表明：1997年亚洲金融风暴对房价的影响大于2008年金融危机；外部经济体的救市政策间接地影响香港房价；在经济不景气的大环境下“孙九招”政策没有立即见效；资本投资者入境计划、住房供给调整与按揭贷款调整对房价的影响较为显著；税收调整对房价影响不显著、对交易量影响显著；SARS爆发使住宅价格下降约1%。     

A post curing strategy toward the feasible covalent adaptable networks in polyacrylate latex films

The implementation of covalent adaptable networks (CANs) in general resin system is becoming attractive. In this work, we propose a simple post-curing strategy based on the core-shell structured acrylate latex for the achievement on both the improved general performance and the CANs characteristics in latex films. The building to the CANs was relied on the introduction of 4,4′-diaminophenyl disulfide as the curing agent, which cured the acetoacetoxy decorated shell polymer through the ketoamine reaction. The metathesis reaction of aromatic disulfides in the crosslinking segments enabled the thermally induced dynamic behavior of the network as revealed in the stress relaxation tests by comparison with other diamine crosslinking agents without the incorporation of disulfide. The synergism of the dynamic crosslinking of the shell polymer and static crosslinking in the core polymer contributed to the improved mechanical strength (15 MPa, strain% = 250%) and the suppressed water adsorption (~1% in 24 h of soaking) of the latex film, which exhibited above 90% of recovery in both strength and strain from a cut-off film damage within 1 h at 80°C. Moreover, the cured latex film could be recycled, and 75% of the mechanical performance was regained after three fragmentation-hot-pressing cycles. These, in addition with the feasible and environmental friendly characteristics, suggest a sustainable paradigm toward the smart thermosetting latex polymers.     

三维位势问题的梯度边界积分方程的新解法

三维位势问题的边界元分析中,关于坐标变量的边界位势梯度的计算是一个困难的问题. 已有一些方法着手解决这个问题,然而,这些方法需要复杂的理论推导和大量的数值计算. 本文提出求解一般边界位势梯度边界积分方程的辅助边值问题法. 该方法构造了与原边界值问题具有相同解域的辅助边值问题,该辅助边值问题具有已知解,因此通过求解此辅助边值问题,可获得梯度边界积分方程对应的系统矩阵,然后将此系统矩阵应用于求解原边值问题,求解过程非常简单,只需求解一个线性系统即可获得原边值问题的解. 值得注意的是,在求解原边值问题时,不再需要重新计算系统矩阵,因此辅助边值问题法的效率并不很差. 辅助边值问题法避免了强奇异积分的计算,具有数学理论简单、程序设计容易、计算精度高等优点,为坐标变量梯度边界积分方程的求解提供了一个新的途径. 3个标准的数值算例验证了方法的有效性.      

Sentinel-2卫星的多光谱轻量级船舶目标检测算法

近年来深度卷积神经网络在可见光船舶检测方面取得了显著的进展，然而，大多数相关研究是通过改进大型的网络结构来提高检测性能，因此加大了对更高计算机性能的需求。此外，可见光图像难以在云、雾、海杂波、黑夜等复杂场景检测到船舶。针对以上问题，提出了一种融合红（red, R）、绿（green, G）、蓝（blue, B）和近红外（NIR）4个波段光谱信息的由粗到精细的轻量型船舶检测算法。与现有的方法中根据光谱特性利用水体检测算法提取水体区域不同之处是该算法是利用改进的水体检测算法来提取船舶候选区域。为获取更准确的候选区域，对船舶、厚云、薄云、平静海面、杂波海面5种场景中4个波段的像素值进行了统计分析，选取近红外大于阈值作为辅助判断，并以其中心点获取候选区域32×32大小的切片，并对切片进行非极大值抑制，由此获得了船舶粗检测结果。随后构建了轻量级LSGFNet网络对船舶候选区域切片进行精细识别。构建的网络融合了1×1卷积提取的波谱特征与3×3的提取几何特征，为防止光谱特征与几何特征的信息在融合时“信息不流通”，在LSGFNet网络中引入了ShuffleNet中的通道打乱机制，并减小了模型结构，与典型的轻量级网络相比具有更好的效果且模型较小。最后，利用Sentinel-2卫星多光谱10 m分辨率数据构建了512×512大小的1 120组数据进行粗检测，以及32×32大小的6 014组数据进行精细网络训练，其中候选区域粗提取的查全率为98.99%，精细识别网络精确度为96.04%，不同场景下的平均精确度为92.98%。实验表明该算法在抑制云层、海浪杂波等干扰的复杂背景下具有较高的检测效率，且训练时间短、计算机性能需求低。     

基于串联两阶段网络DEA模型的供应链规模收益分析

数据包络分析(DEA)是评价供应链系统(Supply chain system)间相对有效性的一种重要的工具,但是传统的DEA不考虑供应链的内部结构,对系统效率评价偏高;而本文所研究两阶段串联供应链系统,考虑把部分中间产品作为最终产品输出,增加额外中间投入的情形.基于所提出的供应链系统结构,本文建立相应的串联结构下的网络DEA模型,并针对所建立模型进行相关理论的研究,给出了串联结构下的生产可能集和规模收益情况判定方法.最后,进行数值实验,以验证我们提出的结论.