是否可以保护张量流训练代码
现在，非常高功率共享的GPU可以以非常便宜的价格进行深度学习培训。但是，保护您的培训数据集和代码是问题，因为GPU所有者可以看到您的文件。我知道，不可能完全防止代码和数据进行反向工程，但我想让它非常困难。我正在考虑在进食模型之前对数据集图像进行加密和解密，但问题是，当我向模型馈送时，任何人都足够聪明地看到我的解密数据。 隐藏TensorFlow代码并构建C ++二进制的选项是什么？我尝试了Nuitka，它不使用TensorFlow。我搜索了Cython也无法正常工作。有TF加密的，但是，似乎与较旧版本的TensorFlow（1.x）一起使用。请提出一些有效的聪明方法。
Now a days very high power shared GPUs are available for deep learning training at very cheap price. But protecting your training dataset and code is problem there because GPU owner can see your files. I know that it is impossible to completely prevent, reverse engineering of the code and data, but I want to make it very very hard. I am thinking to encrypt my datasets images and decrypt problematically before feeding to the model.But the problem is, anyone intelligent enough can see my decrypted data when i feed to the model.
What are the options to hide TensorFlow code and build c++ binary? I tried Nuitka, its not working with tensorflow. I searched that Cython will also not work. There is tf-encrypted but, it seems like that work with older version of tensorflow (1.x). Please suggest some clever way which works.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论