产生式对抗网络(generative adversarial networks,简称GANs)可以生成逼真的图像,因此最近被广泛研究.值得注意的是,概率图生成对抗网络(graphical-GAN)将贝叶斯网络引入产生式对抗网络框架,以无监督的方式学习到数据的隐藏结构.提出了条件概率图生成对抗网络(conditional graphical-GAN),它可以在弱监督环境下,利用粗粒度监督信息来学习到更精细而复杂的结构.条件概率图生成对抗网络的推理和学习遵循与graphical-GAN类似的方法.提出了条件概率图生成对抗网络的两个实例.条件高斯混合模型(conditional Gaussian mixture GAN,简称cGMGAN)可以在给出粗粒度标签的情况下从混合数据中学习细粒度聚类.条件状态空间模型(conditional state space GAN,简称cSSGAN)可以在给定对象标签的情况下学习具有多个对象的视频的动态过程.
A novel flash memory cell with stacked structure （Si substrate/SiGe quantum dots/tunneling oxide/polySi floating gate） is proposed and demonstrated to achieve enhanced F-N tunneling for both programming and erasing. Simulation results indicate the new structure provides high speed and reliability. Experimental results show that the operation voltage can be as much as 4V less than that of conventional full F-N tunneling NAND memory cells. Memory cells with the proposed structure can achieve higher speed, lower voltage, and higher reliability.
