OK. FFFFF**k! 这一篇是大佬《Renyi DP》的参考文献5，说是提供了更生动的比喻？为什么看不懂人家生动的比喻。。
事实上《RDP》这篇在第二部分介绍DP的时候就不知怎么没头没尾的冒出两种场景。。 大佬是这么写的
… We conclude that $(\epsilon,\delta)$-DP is a qualitatively different definition than pure $\epsilon$-DP (unless, of course, $\delta=0$, which we assume not to be the case through the rest of this section).
然后下一段是这样的开头
Even for the simple case of exactly two input databases (such as when the adversary knows the entire dataset except whether it contains a particular record), the $\delta$ additive term encompasses two very different modes in which privacy may fail. In both scenarios $\epsilon$-DP holds with probability $1-\delta$, they differ in what happens with the remaining probability $\delta$…. 这是哪两种场景啊，接着往下就是一种会使privacy degrades gracefully，例子：$\epsilon_1$-DP with probability $\delta/2$, $\epsilon_2$-DP with P $\delta/4$…;另一种是直接以$\delta$的概率泄露某条记录是否为数据库的一部分 – becomes completely exposed.
用我们班同学科学家的话来说就是“放什么屁”。但是科学家应该不会在看论文的时候产生这种疑问吧。。 这种东西看不懂，看这类论文似乎就像看了一遍文字然后并没有什么实质性的感觉，扣半天字眼又让效率变得颇低。 恶性循环恶性循环。。换个坑吧T.T
