skip to main content
US FlagAn official website of the United States government
dot gov icon
Official websites use .gov
A .gov website belongs to an official government organization in the United States.
https lock icon
Secure .gov websites use HTTPS
A lock ( lock ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

Explore Research Products in the PAR
It may take a few hours for recently added research products to appear in PAR search results.


Title: Evaluating Differentially Private Machine Learning in Practice
Differential privacy is a strong notion for privacy that can be used to prove formal guarantees, in terms of a privacy budget, ϵ, about how much information is leaked by a mechanism. When used in privacy-preserving machine learning, the goal is typically to limit what can be inferred from the model about individual training records. However, the calibration of the privacy budget is not well understood. Implementations of privacy-preserving machine learning often select large values of ϵ in order to get acceptable utility of the model, with little understanding of the impact of such choices on meaningful privacy. Moreover, in scenarios where iterative learning procedures are used, relaxed definitions of differential privacy are often used which appear to reduce the needed privacy budget but present poorly understood trade-offs between privacy and utility. In this paper, we quantify the impact of these choices on privacy in experiments with logistic regression and neural network models. Our main finding is that there is no way to obtain privacy for free---relaxed definitions of differential privacy that reduce the amount of noise needed to improve utility also increase the measured privacy leakage. Current mechanisms for differentially private machine learning rarely offer acceptable utility-privacy trade-offs for complex learning tasks: settings that provide limited accuracy loss provide little effective privacy, and settings that provide strong privacy result in useless models.  more » « less
Award ID(s):
1717950
PAR ID:
10099285
Author(s) / Creator(s):
;
Date Published:
Journal Name:
USENIX Security Symposium
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; (, ACSAC 2020: Annual Computer Security Applications Conference)
    null (Ed.)
    Spam phone calls have been rapidly growing from nuisance to an increasingly effective scam delivery tool. To counter this increasingly successful attack vector, a number of commercial smartphone apps that promise to block spam phone calls have appeared on app stores, and are now used by hundreds of thousands or even millions of users. However, following a business model similar to some online social network services, these apps often collect call records or other potentially sensitive information from users’ phones with little or no formal privacy guarantees. In this paper, we study whether it is possible to build a practical collaborative phone blacklisting system that makes use of local differential privacy (LDP) mechanisms to provide clear privacy guarantees. We analyze the challenges and trade-offs related to using LDP, evaluate our LDP-based system on real-world user-reported call records collected by the FTC, and show that it is possible to learn a phone blacklist using a reasonable overall privacy budget and at the same time preserve users’ privacy while maintaining utility for the learned blacklist. 
    more » « less
  2. ; ; ; ; ; (, Lecture notes of the Institute for Computer Sciences Social Informatics and Telecommunications Engineering)
    While embracing various machine learning techniques to make effective decisions in the big data era, preserving the privacy of sensitive data poses significant challenges. In this paper, we develop a privacy-preserving distributed machine learning algorithm to address this issue. Given the assumption that each data provider owns a dataset with different sample size, our goal is to learn a common classifier over the union of all the local datasets in a distributed way without leaking any sensitive information of the data samples. Such an algorithm needs to jointly consider efficient distributed learning and effective privacy preservation. In the proposed algorithm, we extend stochastic alternating direction method of multipliers (ADMM) in a distributed setting to do distributed learning. For preserving privacy during the iterative process, we combine differential privacy and stochastic ADMM together. In particular, we propose a novel stochastic ADMM based privacy-preserving distributed machine learning (PS-ADMM) algorithm by perturbing the updating gradients, that provide differential privacy guarantee and have a low computational cost. We theoretically demonstrate the convergence rate and utility bound of our proposed PS-ADMM under strongly convex objective. Through our experiments performed on real-world datasets, we show that PS-ADMM outperforms other differentially private ADMM algorithms under the same differential privacy guarantee. 
    more » « less
  3. ; ; ; ; (, Journal of Privacy and Confidentiality)
    Differential privacy is a widely accepted formal privacy definition that allows aggregate information about a dataset to be released while controlling privacy leakage for individuals whose records appear in the data. Due to the unavoidable tension between privacy and utility, there have been many works trying to relax the requirements of differential privacy to achieve greater utility.One class of relaxation, which is gaining support outside the privacy community is embodied by the definitions of individual differential privacy (IDP) and bootstrap differential privacy (BDP). Classical differential privacy defines a set of neighboring database pairs and achieves its privacy guarantees by requiring that each pair of neighbors should be nearly indistinguishable to an attacker. The privacy definitions we study, however, aggressively reduce the set of neighboring pairs that are protected.To a non-expert, IDP and BDP can seem very appealing as they echo the same types of privacy explanations that are associated with differential privacy, and also experimentally achieve dramatically better utility. However, we show that they allow a significant portion of the dataset to be reconstructed using algorithms that have arbitrarily low privacy loss under their privacy accounting rules.With the non-expert in mind, we demonstrate these attacks using the preferred mechanisms of these privacy definitions. In particular, we design a set of queries that, when protected by these mechanisms with high noise settings (i.e., with claims of very low privacy loss), yield more precise information about the dataset than if they were not protected at all. The specific attacks here can be defeated and we give examples of countermeasures. However, the defenses are either equivalent to using differential privacy or to ad-hoc methods tailored specifically to the attack (with no guarantee that they protect against other attacks). Thus, the defenses emphasize the deficiencies of these privacy definitions. 
    more » « less
  4. ; ; ; ; ; ; ; ; (, 2018 17th IEEE International Conference on Trust, Security and Privacy In Computing And Communications/12th IEEE International Conference on Big Data Science and Engineering (TrustCom/BigDataSE))
    Preserving privacy in machine learning on multi-party data is of importance to many domains. In practice, existing solutions suffer from several critical limitations, such as significantly reduced utility under privacy constraints or excessive communication burden between the information fusion center and local data providers. In this paper, we propose and implement a new distributed deep learning framework that addresses these shortcomings and preserves privacy more efficiently than previous methods. During the stochastic gradient descent training of a deep neural network, we focus on the parameters with large absolute gradients in order to save privacy budget consumption. We adopt a generalization of the Report-Noisy-Max algorithm in differential privacy to select these gradients and prove its privacy guarantee rigorously. Inspired by the recent novel idea of Terngrad, we also quantize the released gradients to ternary levels {-B, 0, B}, where B is the bound of gradient clipping. Applying Terngrad can significantly reduce the communication cost without incurring severe accuracy loss. Furthermore, we evaluate the performance of our method on a real-world credit card fraud detection data set consisting of millions of transactions. 
    more » « less
  5. ; ; ; ; (, Proceedings on Privacy Enhancing Technologies)
    Abstract Organizations often collect private data and release aggregate statistics for the public’s benefit. If no steps toward preserving privacy are taken, adversaries may use released statistics to deduce unauthorized information about the individuals described in the private dataset. Differentially private algorithms address this challenge by slightly perturbing underlying statistics with noise, thereby mathematically limiting the amount of information that may be deduced from each data release. Properly calibrating these algorithms—and in turn the disclosure risk for people described in the dataset—requires a data curator to choose a value for a privacy budget parameter, ɛ . However, there is little formal guidance for choosing ɛ , a task that requires reasoning about the probabilistic privacy–utility tradeoff. Furthermore, choosing ɛ in the context of statistical inference requires reasoning about accuracy trade-offs in the presence of both measurement error and differential privacy (DP) noise. We present Vi sualizing P rivacy (ViP), an interactive interface that visualizes relationships between ɛ , accuracy, and disclosure risk to support setting and splitting ɛ among queries. As a user adjusts ɛ , ViP dynamically updates visualizations depicting expected accuracy and risk. ViP also has an inference setting, allowing a user to reason about the impact of DP noise on statistical inferences. Finally, we present results of a study where 16 research practitioners with little to no DP background completed a set of tasks related to setting ɛ using both ViP and a control. We find that ViP helps participants more correctly answer questions related to judging the probability of where a DP-noised release is likely to fall and comparing between DP-noised and non-private confidence intervals. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email