A distribution class exists that is learnable non-privately in TV distance with finite samples but not under differential privacy, weakly refuting Ashtiani's conjecture.
A bias-variance-privacy trilemma for statistical estimation
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
verdicts
UNVERDICTED 2representative citing papers
Multidimensional simplex transformations on [0,1]-bounded variables extend the free lunch for private dataset size estimation, refining sufficient statistics for differentially private simple linear regression via OLS with claimed analytical and numerical gains.
citing papers explorer
-
Not All Learnable Distribution Classes are Privately Learnable
A distribution class exists that is learnable non-privately in TV distance with finite samples but not under differential privacy, weakly refuting Ashtiani's conjecture.
-
Refined Differentially Private Linear Regression via Extension of a Free Lunch Result
Multidimensional simplex transformations on [0,1]-bounded variables extend the free lunch for private dataset size estimation, refining sufficient statistics for differentially private simple linear regression via OLS with claimed analytical and numerical gains.