To guide the illuminating design to improve the on-state performances of gallium arsenide(GaAs)photoconductive semiconductor switch(PCSS),the effect of spot size on the operation mode of GaAsPCSS based on a semi-insul...To guide the illuminating design to improve the on-state performances of gallium arsenide(GaAs)photoconductive semiconductor switch(PCSS),the effect of spot size on the operation mode of GaAsPCSS based on a semi-insulating wafer with a thickness of 1 mm,triggered by a 1064-nm extrinsic laser beam with the rectangular spot,has been investigated experimentally.It is found that the variation of the spot size in length and width can act on the different parts of the output waveform integrating the characteristics of the linear and nonlinear modes,and then significantly boosts the PCSS toward different operation modes.On this basis,a two-channel model containing the active and passive parts is introduced to interpret the relevant influencing mechanisms.Results indicate that the increased spot length can peak the amplitude of static domains in the active part to enhance the development of the nonlinear switching,while the extended spot width can change the distribution of photogenerated carriers on both parts to facilitate the linear switching and weaken the nonlinear switching,which have been proved by comparing the domain evolutions under different spot sizes.展开更多
受对抗样本自身可迁移属性的影响,传统对抗样本防御方法的防御效果存在不稳定的情况,为此,提出基于深度学习的对抗样本防御方法。文章借助深度学习算法,构建了对抗样本伊辛模型,设置模型的初始状态为神经网络的输入数据,采用自旋状态表...受对抗样本自身可迁移属性的影响,传统对抗样本防御方法的防御效果存在不稳定的情况,为此,提出基于深度学习的对抗样本防御方法。文章借助深度学习算法,构建了对抗样本伊辛模型,设置模型的初始状态为神经网络的输入数据,采用自旋状态表示每一个神经元值与对抗样本伊辛模型的格点,并利用神经网络中卷积运算的特征,消解势场中预先给定的外部磁化作用,以最大限减少低对抗样本伊辛模型在能量作用下的局部自旋问题。在对抗样本防御阶段,利用对抗样本伊辛模型的通道相关性,生成重要性掩码对通道的激活进行调整,并结合对抗样本伊辛模型通道梯度累积值的实际情况设置了差异化的重要性掩码生成函数。在应用测试过程中,为验证防御效果,在快速梯度下降法(Fast Gradient Sign Method,FGSM)、Deepfool、C&W(Carlini and Wagner)攻击算法、投影梯度下降(Projected Gradient Descent,PFD)、集成对抗检测器(Energy-Aware Data-centric,EAD)共5种对抗策略下设计了对抗样本防御方法,对比不同对抗样本防御方法的性能,发现文章提出的基于深度学习的对抗样本防御方法的曲线下的面积(Area Under the Curve,AUC)值稳定在0.95以上,说明对抗样本防御方法具有较好的防御性能。展开更多
基金supported in part by the Huxiang Youth Talent Support Program(No.2020RC3030)in part by the Foundation of State Key Laboratory of Pulsed Power Laser Technology(Nos.SKL2021ZR02 and SKL2021KF05)。
文摘To guide the illuminating design to improve the on-state performances of gallium arsenide(GaAs)photoconductive semiconductor switch(PCSS),the effect of spot size on the operation mode of GaAsPCSS based on a semi-insulating wafer with a thickness of 1 mm,triggered by a 1064-nm extrinsic laser beam with the rectangular spot,has been investigated experimentally.It is found that the variation of the spot size in length and width can act on the different parts of the output waveform integrating the characteristics of the linear and nonlinear modes,and then significantly boosts the PCSS toward different operation modes.On this basis,a two-channel model containing the active and passive parts is introduced to interpret the relevant influencing mechanisms.Results indicate that the increased spot length can peak the amplitude of static domains in the active part to enhance the development of the nonlinear switching,while the extended spot width can change the distribution of photogenerated carriers on both parts to facilitate the linear switching and weaken the nonlinear switching,which have been proved by comparing the domain evolutions under different spot sizes.
文摘受对抗样本自身可迁移属性的影响,传统对抗样本防御方法的防御效果存在不稳定的情况,为此,提出基于深度学习的对抗样本防御方法。文章借助深度学习算法,构建了对抗样本伊辛模型,设置模型的初始状态为神经网络的输入数据,采用自旋状态表示每一个神经元值与对抗样本伊辛模型的格点,并利用神经网络中卷积运算的特征,消解势场中预先给定的外部磁化作用,以最大限减少低对抗样本伊辛模型在能量作用下的局部自旋问题。在对抗样本防御阶段,利用对抗样本伊辛模型的通道相关性,生成重要性掩码对通道的激活进行调整,并结合对抗样本伊辛模型通道梯度累积值的实际情况设置了差异化的重要性掩码生成函数。在应用测试过程中,为验证防御效果,在快速梯度下降法(Fast Gradient Sign Method,FGSM)、Deepfool、C&W(Carlini and Wagner)攻击算法、投影梯度下降(Projected Gradient Descent,PFD)、集成对抗检测器(Energy-Aware Data-centric,EAD)共5种对抗策略下设计了对抗样本防御方法,对比不同对抗样本防御方法的性能,发现文章提出的基于深度学习的对抗样本防御方法的曲线下的面积(Area Under the Curve,AUC)值稳定在0.95以上,说明对抗样本防御方法具有较好的防御性能。
