Differentially Private Federated Quantiles with the Distributed Discrete Gaussian Mechanism - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Differentially Private Federated Quantiles with the Distributed Discrete Gaussian Mechanism

Résumé

The computation of analytics in a federated environment plays an increasingly important role in data science and machine learning. We consider the differentially private computation of the quantiles of a distribution of values stored on a population of clients. We present two quantile estimation algorithms based on the distributed discrete Gaussian mechanism that are compatible with secure aggregation. Based on a privacy-utility analysis and numerical experiments, we delineate the regime under which each one is superior. We find that the algorithm with suboptimal asymptotic performance works the best on moderate problem sizes typical in federated learning with client sampling. We apply these algorithms to augment distributionally robust federated learning with differential privacy.
Fichier principal
Vignette du fichier
main.pdf (679.02 Ko) Télécharger le fichier
comp_128_32_chisq.pdf (20.18 Ko) Télécharger le fichier
comp_512_64_uniform.pdf (20.54 Ko) Télécharger le fichier
differential_privacy.pdf (282.06 Ko) Télécharger le fichier
dp-flat-v-heir_0.25.pdf (22.14 Ko) Télécharger le fichier
dp-flat-v-heir_0.5.pdf (22.12 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03834472 , version 1 (30-10-2022)

Licence

Paternité

Identifiants

  • HAL Id : hal-03834472 , version 1

Citer

Krishna Pillutla, Yassine Laguel, Jérôme Malick, Zaid Harchaoui. Differentially Private Federated Quantiles with the Distributed Discrete Gaussian Mechanism. International Workshop on Federated Learning: Recent Advances and New Challenges, Dec 2022, New Orleans, United States. ⟨hal-03834472⟩
72 Consultations
88 Téléchargements

Partager

Gmail Facebook X LinkedIn More