RT @XiaoLiMeng1: Fully agree since no new values are created by sampling. But subsampling can enhance DP. A lot more to be investigated; t…
4,649 followers
3,196 followers
Fully agree since no new values are created by sampling. But subsampling can enhance DP. A lot more to be investigated; there are interesting/tough questions to answer — under what conditions it is the right tradeoff for prioritizing outliers over most st
2,925 followers
@XiaoLiMeng1 @rnishimura My understanding is that sampling does not necessarily satisfy differential privacy definitions (at least not the strict ɛ-DP, and not in a nonparametric sense that it would apply to any variable, like the design-based variance inf
1,037 followers