【文献阅读】Local and Central Differential Privacy in FL

论文标题：Local and Central Differential Privacy for Robustness and Privacy in Federated Learning
关键词：Local Differential Privacy；Central Differential Privacy；FL

类似综述+实验论文，主要讲了在 FL 场景下 Local Differential Privacy （LDP）和 Central Differential Privacy （CDP）在遭受推断攻击（针对privacy ） 和 投毒攻击（针对robustness） 时的保护能力。

定义

Local Differential Privacy

主要： protect an individual’s data

Central Differential Privacy

主要：guarantees that the output of the aggregation function is indistinguishable，但是要确保 server 是可信的

主要结论

Our main intuition is that CDP limits the information exposed about a specific participant, while LDP does so for records in a participant’s dataset;