The most famous defense against model theft attacks is PATE (https://arxiv.org/abs/1802.08908), a privacy framework developed by Ian Goodfellow et al. As shown in the above figure, the key idea behind PATE is to divide the data and train multiple combined models to make decisions. This decision was then masked by noise from other different privacy systems.
也不吃肉，把骨头给他啃就成了。
< p > This is a paragraph.
