Convolutional Neural Network Protection Method of Lenet-5-Like Structure
Convolutional Neural Network Protection Method of Lenet-5-Like Structure
复制标题
Lenet-5类结构的卷积神经网络保护方法
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
Leyu Dai
中科院分区：
文献类型：
--
作者：
Lei Sun;Yuehan Wang;Leyu Dai
在本文中，我们建立和描述了一个卷积神经网络的Lenet-5型结构的保护方法。通过在前向传播算法的不同层中加入保护锁模块，使未授权用户无法使用受保护的类Lenet-5结构的卷积神经网络，从而得到受保护的卷积神经网络。实验结果表明，在第一层和中间层增加保护锁模块可以获得最佳的保护效果。授权用户在添加解锁模块后，可以使用受保护的Lenet-5类结构卷积神经网络。通过增加保护锁定模块和解锁模块，不影响解锁卷积神经网络的准确性。在Lenet-5类结构的卷积神经网络上增加保护锁模块实现保护。
In this paper, we build and describe a convolutional neural network protection method of Lenet-5-like structure. The protected convolutional neural network is obtained by adding the protection lock module in forward propagation algorithm in different layers so that unauthorized user can not use protected convolutional neural network of Lenet-5-like structure. The experimental results show that protection lock module be added in first and middle layers would gain the best protection effect. Authorized users can use protected convolutional neural network of Lenet-5-like structure after adding unlocking module. The accuracy of unlocked convolutional neural network is not affected by adding protection lock module and unlocking module. To add protection lock module realize protection on the convolutional neural network of Lenet-5-like structure.