Private selection from private candidates

Differentially Private algorithms often need to select the best amongst many candidate options. Classical works on this selection problem require that the candidates’ goodness, measured as a real-valued score function, does not change by much when one person’s data changes. In many applications such as hyperparameter optimization, this stability assumption is much too strong. In this work, we consider the selection problem under a much weaker stability assumption on the candidates, namely that the score functions are differentially private. Under this assumption, we present algorithms that are near-optimal along the three relevant dimensions: privacy, utility and computational efficiency. Our result can be seen as a generalization of the exponential mechanism and its existing generalizations. We also develop an online version of our algorithm, that can be seen as a generalization of the sparse vector technique to this weaker stability assumption. We show how our results imply better algorithms for hyperparameter selection in differentially private machine learning, as well as for adaptive data analysis.

[1]  Kamalika Chaudhuri,et al.  A Stability-based Validation Procedure for Differentially Private Machine Learning , 2013, NIPS.

[2]  Cynthia Dwork,et al.  Calibrating Noise to Sensitivity in Private Data Analysis , 2006, TCC.

[3]  Kunal Talwar,et al.  Mechanism Design via Differential Privacy , 2007, 48th Annual IEEE Symposium on Foundations of Computer Science (FOCS'07).

[4]  Adam D. Smith,et al.  Privacy-preserving statistical estimation with optimal convergence rates , 2011, STOC '11.

[5]  Stephen E. Fienberg,et al.  Privacy-Preserving Data Sharing for Genome-Wide Association Studies , 2012, J. Priv. Confidentiality.

[6]  Kobbi Nissim,et al.  Differentially Private Release and Learning of Threshold Functions , 2015, 2015 IEEE 56th Annual Symposium on Foundations of Computer Science.

[7]  Sofya Raskhodnikova,et al.  Lipschitz Extensions for Node-Private Graph Statistics and the Generalized Exponential Mechanism , 2016, 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS).

[8]  Sofya Raskhodnikova,et al.  Smooth sensitivity and sampling in private data analysis , 2007, STOC '07.

[9]  Toniann Pitassi,et al.  Preserving Statistical Validity in Adaptive Data Analysis , 2014, STOC.

[10]  Hiroshi Nakagawa,et al.  Differential Privacy without Sensitivity , 2016, NIPS.

[11]  Ian Goodfellow,et al.  Deep Learning with Differential Privacy , 2016, CCS.

[12]  Martín Abadi,et al.  Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data , 2016, ICLR.

[13]  Sanjiv Kumar,et al.  Truncated Laplacian Mechanism for Approximate Differential Privacy , 2018, ArXiv.

[14]  Moni Naor,et al.  On the complexity of differentially private data release: efficient algorithms and hardness results , 2009, STOC '09.

[15]  Toniann Pitassi,et al.  The Limits of Two-Party Differential Privacy , 2010, 2010 IEEE 51st Annual Symposium on Foundations of Computer Science.

[16]  A. Gelman,et al.  The statistical crisis in science , 2014 .

[17]  Cynthia Dwork,et al.  Differential privacy and robust statistics , 2009, STOC '09.

[18]  M. Chao,et al.  Negative Moments of Positive Random Variables , 1972 .

[19]  Jing Lei,et al.  Differentially private model selection with penalized and constrained likelihood , 2016, 1607.04204.

[20]  Cynthia Dwork,et al.  Private False Discovery Rate Control , 2015, ArXiv.

[21]  Vikas K. Garg,et al.  Supervising Unsupervised Learning , 2017, NeurIPS.

[22]  Amos Beimel,et al.  Private Learning and Sanitization: Pure vs. Approximate Differential Privacy , 2013, APPROX-RANDOM.

[23]  Aaron Roth,et al.  Differentially private combinatorial optimization , 2009, SODA '10.

[24]  Anand D. Sarwate,et al.  Near-optimal Differentially Private Principal Components , 2012, NIPS.

[25]  Adam D. Smith,et al.  Differentially Private Feature Selection via Stability Arguments, and the Robustness of the Lasso , 2013, COLT.

[26]  Kunal Talwar,et al.  On differentially private low rank approximation , 2013, SODA.

[27]  Ashwin Machanavajjhala,et al.  Pythia: Data Dependent Differentially Private Algorithm Selection , 2017, SIGMOD Conference.

[28]  Rebecca N. Wright,et al.  Differential privacy: an exploration of the privacy-utility landscape , 2013 .

[29]  Seth Neel,et al.  Accuracy First: Selecting a Differential Privacy Level for Accuracy Constrained ERM , 2017, NIPS.

[30]  Adam D. Smith,et al.  Discovering frequent patterns in sensitive data , 2010, KDD.

[31]  Aaron Roth,et al.  The Algorithmic Foundations of Differential Privacy , 2014, Found. Trends Theor. Comput. Sci..

[32]  Raef Bassily,et al.  Algorithmic stability for adaptive data analysis , 2015, STOC.

[33]  Kamalika Chaudhuri,et al.  The Large Margin Mechanism for Differentially Private Maximization , 2014, NIPS.

[34]  Ameet Talwalkar,et al.  Hyperband: Bandit-Based Configuration Evaluation for Hyperparameter Optimization , 2016, ICLR.

[35]  Sofya Raskhodnikova,et al.  What Can We Learn Privately? , 2008, 2008 49th Annual IEEE Symposium on Foundations of Computer Science.

[36]  Roman Garnett,et al.  Differentially Private Bayesian Optimization , 2015, ICML.