• Corpus ID: 246411320

A Joint Exponential Mechanism For Differentially Private Top-k

@article{Gillenwater2022AJE,
  title={A Joint Exponential Mechanism For Differentially Private Top-k},
  author={Jennifer Gillenwater and Matthew Joseph and Andr{\'e}s Mu{\~n}oz Medina and M{\'o}nica Ribero},
  journal={ArXiv},
  year={2022},
  volume={abs/2201.12333}
}
We present a novel differentially private algorithm for releasing the set of k elements with the highest counts from a data domain of d elements. We define a “joint” instance of the exponential mechanism (EM) whose output space consists of all O ( d k ) size- k subsets; yet, we are able to show how to sample from this EM in only time ˜ O ( dk 3 ) . Experiments suggest that this joint approach can yield utility improvements over the existing state of the art for small problem sizes. 

Figures from this paper

Differentially Private Top-k Selection via Canonical Lipschitz Mechanism
TLDR
This work presents an additive noise mechanism with a single DP-proof via a mandated Lipschitz property for the noise distribution, and a composition-free approach to subset selection that improves utility guarantees by an Ω(log k) factor compared to oneby-one selection via sequential composition.
Adaptive Private-K-Selection with Adaptive K and Application to Multi-label PATE
TLDR
This work proposes to privately release a data-dependent choice of k such that the gap between k -th and the ( k + 1) st “quality” is large and improves the privacy-utility trade-ofis compared to the previous top- k selection algorithms theoretically and empirically.
Histogram Estimation under User-level Privacy with Heterogeneous Data
TLDR
This work proposes an algorithm based on a clipping strategy that almost achieves a two-approximation with respect to the best clipping threshold in hindsight and proves that the clipping bias can be significantly reduced when the counts are from non-i.i.d. Poisson distributions.

References

SHOWING 1-10 OF 32 REFERENCES
Permute-and-Flip: A new mechanism for differentially private selection
TLDR
This work proposes a new mechanism for differentially private selection based on a careful analysis of the privacy constraints, which can offer improvements up to a factor of two and runs in linear time.
Oneshot Differentially Private Top-k Selection
TLDR
The oneshot Laplace mechanism, which generalizes the wellknown Report Noisy Max mechanism to reporting noisy top-k elements, is presented and it is shown that the oneshotLaplace mechanism with a noise level of Õ( √ k/ε) is approximately differentially private.
Practical Differentially Private Top-k Selection with Pay-what-you-get Composition
TLDR
This work designs algorithms that ensures (approximate) $(\epsilon,\delta>0)$-differential privacy and only needs access to the true top-$\bar{k}$ elements from the data for any chosen $\bar{ k} \geq k$.
Tight Lower Bounds for Differentially Private Selection
TLDR
The fingerprinting method is used to provide tight lower bounds for answering an entire set of d queries, but often only some much smaller set of k queries are relevant and the extension allows us to prove lower bounds that depend on both the number of relevant queries and the total number of queries.
Duff: A Dataset-Distance-Based Utility Function Family for the Exponential Mechanism
TLDR
D Duff is an affirmative answer to the open question of whether it is possible to have a noise distribution whose variance is proportional to smooth sensitivity and whose tails decay at a faster-than-polynomial rate.
Fingerprinting Codes and the Price of Approximate Differential Privacy
TLDR
The results rely on the existence of short fingerprinting codes (Boneh and Shaw, CRYPTO'95; Tardos, STOC'03), which are closely connected to the sample complexity of differentially private data release.
Free Gap Information from the Differentially Private Sparse Vector and Noisy Max Mechanisms
TLDR
This paper shows that Noisy Max and Sparse Vector can release additional information for free (i.e., at no additional privacy cost) and can improve the accuracy of certain subsequent counting queries by up to 50%.
On the geometry of differential privacy
TLDR
The lower bound is strong enough to separate the concept of differential privacy from the notion of approximate differential privacy where an upper bound of O(√{d}/ε) can be achieved.
Top-k frequent itemsets via differentially private FP-trees
TLDR
An approach that first identifies top-k frequent itemsets, then uses them to construct a compact, differentially private FP-tree, which gives substantially better results than prior approaches, especially for high levels of privacy.
The Price of Selection in Differential Privacy
TLDR
The lower bound is the first to show that selecting the largest columns requires more data than simply estimating the value of those $k$ columns, which can be done using a dataset of size just $n \gtrsim k$.
...
...