Differentially private false discovery rate control

Dwork, Cynthia; Su, Weijie; Zhang, Li

doi:10.29012/jpc.755

Cited by 7 publications

(3 citation statements)

References 48 publications

(77 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Laplace or Gaussian noise) to the test statistic or, more generally, the queried result (Dwork et al, 2015a). While theoretical guarantees are available for differential privacy based methods for test data reuse (e.g., (Dwork et al, 2015b;Russo and Zou, 2016;Rogers et al, 2016;Cummings et al, 2016;Dwork et al, 2017;Steinke, 2017, 2018;Shenfeld and Ligett, 2019;Gossmann et al, 2021) and others), the required size of the test dataset is prohibitively large for many application domains or require injecting very large amounts of noise (Rogers et al, 2019;Gossmann et al, 2021). An alternative approach is to directly limit the number of bits of information released to the model developer by discretizing the queried result along some grid (Blum and Hardt, 2015).…”

Section: Related Workmentioning

confidence: 99%

Sequential algorithmic modification with test data reuse

Feng¹,

Pennello²,

Petrick³

et al. 2022

Preprint

View full text Add to dashboard Cite

After initial release of a machine learning algorithm, the model can be fine-tuned by retraining on subsequently gathered data, adding newly discovered features, or more. Each modification introduces a risk of deteriorating performance and must be validated on a test dataset. It may not always be practical to assemble a new dataset for testing each modification, especially when most modifications are minor or are implemented in rapid succession. Recent works have shown how one can repeatedly test modifications on the same dataset and protect against overfitting by (i) discretizing test results along a grid and (ii) applying a Bonferroni correction to adjust for the total number of modifications considered by an adaptive developer. However, the standard Bonferroni correction is overly conservative when most modifications are beneficial and/or highly correlated. This work investigates more powerful approaches using alpha-recycling and sequentially-rejective graphical procedures (SRGPs). We introduce novel extensions that account for correlation between adaptively chosen algorithmic modifications. In empirical analyses, the SRGPs control the error rate of approving unacceptable modifications and approve a substantially higher number of beneficial modifications than previous approaches.

show abstract

Section: Related Workmentioning

confidence: 99%

Sequential algorithmic modification with test data reuse

Feng¹,

Pennello²,

Petrick³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…This means that differentially-private OLS based on these ERM algorithms requires us to devise new versions of these algorithms, making this a second step in this line of work... (After first understanding what we can do using existing algorithms.) We leave this approach -as well as performing private hypothesis testing using a PTR-type algorithm (Dwork & Lei, 2009) (output merely reject / don't-reject decision without justification), or releasing only relevant tests judging by their p-values (Dwork et al, 2015) -for future work.…”

Section: Introductionmentioning

confidence: 99%

“…Discussion. Some works have already looked at the intersection of differentially privacy and statistics (Dwork & Lei, 2009;Smith, 2011;Chaudhuri & Hsu, 2012;Duchi et al, 2013;Dwork et al, 2015) (especially focusing on robust statistics and rate of convergence). But only a handful of works studied the significance and power of hypotheses testing under differential privacy, without arguing that the noise introduced by differential privacy vanishes asymptotically (Vu & Slavkovic, 2009;Uhler et al, 2013;Wang et al, 2015;Rogers et al, 2016).…”

Section: Introductionmentioning

confidence: 99%

Differentially Private Ordinary Least Squares

Sheffet

2019

JPC

View full text Add to dashboard Cite

Linear regression is one of the most prevalent techniques in machine learning; however, it is also common to use linear regression for its explanatory capabilities rather than label prediction. Ordinary Least Squares (OLS) is often used in statistics to establish a correlation between an attribute (e.g. gender) and a label (e.g. income) in the presence of other (potentially correlated) features. OLS assumes a particular model that randomly generates the data, and derives tvalues -representing the likelihood of each real value to be the true correlation. Using t-values, OLS can release a confidence interval, which is an interval on the reals that is likely to contain the true correlation; and when this interval does not intersect the origin, we can reject the null hypothesis as it is likely that the true correlation is non-zero. Our work aims at achieving similar guarantees on data under differentially private estimators. First, we show that for wellspread data, the Gaussian Johnson-Lindenstrauss Transform (JLT) gives a very good approximation of t-values; secondly, when JLT approximates Ridge regression (linear regression with l 2 -regularization) we derive, under certain conditions, confidence intervals using the projected data; lastly, we derive, under different conditions, confidence intervals for the "Analyze Gauss" algorithm (Dwork et al., 2014).

show abstract

Survey of Recent Results in Privacy-Preserving Mechanisms for Multi-Agent Systems

Kossek,

Stefanovic

2024

J Intell Robot Syst

View full text Add to dashboard Cite

Privacy-preserving communication in cooperative control is essential for effective operations of various systems where sensitive information needs to be protected. This includes systems such as smart grids, traffic management systems, autonomous vehicle networks, healthcare systems, financial networks, and social networks. Recent privacy-preserving cooperative control literature is categorized and discussed in this paper. Advantages and disadvantages of differential privacy and encryption-based privacy-preserving protocols are described. The objective of this work is to examine and analyze existing research and knowledge related to the preservation of privacy in the context of cooperative control. This paper aims to identify and present a range of approaches, techniques, and methodologies that have been proposed or employed to address privacy concerns in multi-agent systems. It seeks to explore the current challenges, limitations, and gaps in the existing literature. It also aims to consolidate the findings from various studies to provide an overview of privacy-preserving cooperative control in multi-agent systems. The goal is to assist in the development of novel privacy-preserving mechanisms for cooperative control.

show abstract

Differentially private false discovery rate control

Cited by 7 publications

References 48 publications

Sequential algorithmic modification with test data reuse

Sequential algorithmic modification with test data reuse

Differentially Private Ordinary Least Squares

Survey of Recent Results in Privacy-Preserving Mechanisms for Multi-Agent Systems

Contact Info

Product

Resources

About