skip to main content

Attention:

The NSF Public Access Repository (NSF-PAR) system and access will be unavailable from 11:00 PM ET on Thursday, June 13 until 2:00 AM ET on Friday, June 14 due to maintenance. We apologize for the inconvenience.


Title: Differential Privacy Has Disparate Impact on Model Accuracy
Differential privacy (DP) is a popular mechanism for training machine learning models with bounded leakage about the presence of specific points in the training data. The cost of differential privacy is a reduction in the model's accuracy. We demonstrate that in the neural networks trained using differentially private stochastic gradient descent (DP-SGD), this cost is not borne equally: accuracy of DP models drops much more for the underrepresented classes and subgroups. For example, a gender classification model trained using DP-SGD exhibits much lower accuracy for black faces than for white faces. Critically, this gap is bigger in the DP model than in the non-DP model, i.e., if the original model is unfair, the unfairness becomes worse once DP is applied. We demonstrate this effect for a variety of tasks and models, including sentiment analysis of text and image classification. We then explain why DP training mechanisms such as gradient clipping and noise addition have disproportionate effect on the underrepresented and more complex subgroups, resulting in a disparate reduction of model accuracy.  more » « less
Award ID(s):
1700832
NSF-PAR ID:
10249767
Author(s) / Creator(s):
; ;
Date Published:
Journal Name:
Advances in neural information processing systems
Volume:
32
ISSN:
1049-5258
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Differential privacy (DP) is a popular mechanism for training machine learning models with bounded leakage about the presence of specific points in the training data. The cost of differential privacy is a reduction in the model's accuracy. We demonstrate that in the neural networks trained using differentially private stochastic gradient descent (DP-SGD), this cost is not borne equally: accuracy of DP models drops much more for the underrepresented classes and subgroups. For example, a gender classification model trained using DP-SGD exhibits much lower accuracy for black faces than for white faces. Critically, this gap is bigger in the DP model than in the non-DP model, i.e., if the original model is unfair, the unfairness becomes worse once DP is applied. We demonstrate this effect for a variety of tasks and models, including sentiment analysis of text and image classification. We then explain why DP training mechanisms such as gradient clipping and noise addition have disproportionate effect on the underrepresented and more complex subgroups, resulting in a disparate reduction of model accuracy. 
    more » « less
  2. Andreas Krause, Emma Brunskill (Ed.)
    Differentially private (DP) machine learning techniques are notorious for their degradation of model utility (e.g., they degrade classification accuracy). A recent line of work has demonstrated that leveraging public data can improve the trade-off between privacy and utility when training models with DP guaranteed. In this work, we further explore the potential of using public data in DP models, showing that utility gains can in fact be significantly higher than what shown in prior works. Specifically, we introduce DOPE-SGD, a modified DP-SGD algorithm that leverages public data during its training. DOPE-SGD uses public data in two complementary ways: (1) it uses advance augmentation techniques that leverages public data to generate synthetic data that is effectively embedded in multiple steps of the training pipeline; (2) it uses a modified gradient clipping mechanism (which is a standard technique in DP training) to change the origin of gradient vectors using the information inferred from available public and synthetic data, therefore boosting utility. We also introduce a technique to ensemble intermediate DP models by leveraging the post processing property of differential privacy to further improve the accuracy of the predictions. Our experimental results demonstrate the effectiveness of our approach in improving the state-of-the-art in DP machine learning across multiple datasets, network architectures, and application domains. For instance, assuming access to 2,000 public images, and for a privacy budget of 𝜀=2,𝛿=10−5, our technique achieves an accuracy of 75.1 on CIFAR10, significantly higher than 68.1 achieved by the state of the art. 
    more » « less
  3. Larochelle, Hugo ; Hadsell, Raia ; Cho, Kyunghyun (Ed.)
    In deep learning, leveraging transfer learning has recently been shown to be an effective strategy for training large high performance models with Differential Privacy (DP). Moreover, somewhat surprisingly, recent works have found that privately training just the last layer of a pre-trained model provides the best utility with DP. While past studies largely rely on using first-order differentially private training algorithms like DP-SGD for training large models, in the specific case of privately learning from features, we observe that computational burden is often low enough to allow for more sophisticated optimization schemes, including second-order methods. To that end, we systematically explore the effect of design parameters such as loss function and optimization algorithm. We find that, while commonly used logistic regression performs better than linear regression in the non-private setting, the situation is reversed in the private setting. We find that least-squares linear regression is much more effective than logistic regression from both privacy and computational standpoint, especially at stricter epsilon values (ε < 1). On the optimization side, we also explore using Newton’s method, and find that second-order information is quite helpful even with privacy, although the benefit significantly diminishes with stricter privacy guarantees. While both methods use second-order information, least squares is more effective at lower epsilon values while Newton’s method is more effective at larger epsilon values. To combine the benefits of both methods, we propose a novel optimization algorithm called DP-FC, which leverages feature covariance instead of the Hessian of the logistic regression loss and performs well across all ε values we tried. With this, we obtain new SOTA results on ImageNet-1k, CIFAR-100 and CIFAR-10 across all values of ε typically considered. Most remarkably, on ImageNet-1K, we obtain top-1 accuracy of 88% under DP guarantee of (8, 8 ∗ 10−7) and 84.3% under (0.1, 8 ∗ 10−7). 
    more » « less
  4. In differentially private stochastic gradient descent (DPSGD), gradient clipping and random noise addition disproportionately affect underrepresented and complex classes and subgroups. As a consequence, DPSGD has disparate impact: the accuracy of a model trained using DPSGD tends to decrease more on these classes and subgroups vs. the original, non-private model. If the original model is unfair in the sense that its accuracy is not the same across all subgroups, DPSGD exacerbates this unfairness. In this work, we study the inequality in utility loss due to differential privacy, which compares the changes in prediction accuracy w.r.t. each group between the private model and the non-private model. We analyze the cost of privacy w.r.t. each group and explain how the group sample size along with other factors is related to the privacy impact on group accuracy. Furthermore, we propose a modified DPSGD algorithm, called DPSGD-F, to achieve differential privacy, equal costs of differential privacy, and good utility. DPSGD-F adaptively adjusts the contribution of samples in a group depending on the group clipping bias such that differential privacy has no disparate impact on group accuracy. Our experimental evaluation shows the effectiveness of our removal algorithm on achieving equal costs of differential privacy with satisfactory utility. 
    more » « less
  5. The emergence of mobile apps (e.g., location-based services, geo-social networks, ride-sharing) led to the collection of vast amounts of trajectory data that greatly benefit the understanding of individual mobility. One problem of particular interest is next-location prediction, which facilitates location-based advertising, point-of-interest recommendation, traffic optimization,etc. However, using individual trajectories to build prediction models introduces serious privacy concerns, since exact whereabouts of users can disclose sensitive information such as their health status or lifestyle choices. Several research efforts focused on privacy-preserving next-location prediction, but they have serious limitations: some use outdated privacy models (e.g., k-anonymity), while others employ learning models with limited expressivity (e.g., matrix factorization). More recent approaches(e.g., DP-SGD) integrate the powerful differential privacy model with neural networks, but they provide only generic and difficult-to-tune methods that do not perform well on location data, which is inherently skewed and sparse.We propose a technique that builds upon DP-SGD, but adapts it for the requirements of next-location prediction. We focus on user-level privacy, a strong privacy guarantee that protects users regardless of how much data they contribute. Central to our approach is the use of the skip-gram model, and its negative sampling technique. Our work is the first to propose differentially-private learning with skip-grams. In addition, we devise data grouping techniques within the skip-gram framework that pool together trajectories from multiple users in order to accelerate learning and improve model accuracy. Experiments conducted on real datasets demonstrate that our approach significantly boosts prediction accuracy compared to existing DP-SGD techniques. 
    more » « less