Systematic Evaluation of Privacy Risks of Machine Learning Models

Abstract

Machine learning models are prone to memorizing sensitive data, making them vulnerable to membership inference attacks in which an adversary aims to guess if an input sample was used to train the model. In this paper, we show that prior work on membership inference attacks may severely underestimate the privacy risks by relying solely on training custom neural network classifiers to perform attacks and focusing only on the aggregate results over data samples, such as the attack accuracy. To overcome these limitations, we first propose to benchmark membership inference privacy risks by improving existing non-neural network based inference attacks and proposing a new inference attack method based on a modification of prediction entropy. We also propose benchmarks for defense mechanisms by accounting for adaptive adversaries with knowledge of the defense and also accounting for the trade-off between model accuracy and privacy risks. Using our benchmark attacks, we demonstrate that existing defense approaches are not as effective as previously reported. Next, we introduce a new approach for fine-grained privacy analysis by formulating and deriving a new metric called the privacy risk score. Our privacy risk score metric measures an individual sample's likelihood of being a training member, which allows an adversary to identify samples with high privacy risks and perform attacks with high confidence. We experimentally validate the effectiveness of the privacy risk score and demonstrate that the distribution of privacy risk score across individual samples is heterogeneous. Finally, we perform an in-depth investigation for understanding why certain samples have high privacy risks, including correlations with model sensitivity, generalization error, and feature embeddings. Our work emphasizes the importance of a systematic and rigorous evaluation of privacy risks of machine learning models.

arXiv commentAccepted by USENIX Security 2021, code is available at https://github.com/inspire-group/membership-inference-evaluation

References in corpus (8)

• Deep Learning with Differential Privacy
• Comprehensive Privacy Analysis of Deep Learning: Passive and Active White-box Inference Attacks against Centralized and Federated Learning
• Learning Differentially Private Recurrent Language Models
• The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks
• Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data
• Understanding Membership Inferences on Well-Generalized Learning Models
• Stolen Memories: Leveraging Model Memorization for Calibrated White-Box Membership Inference
• Updates-Leak: Data Set Inference and Reconstruction Attacks in Online Learning

Cited by in corpus (8)

• Do Gradient Inversion Attacks Make Federated Learning Unsafe?
• DP-Forward: Fine-tuning and Inference on Language Models with Differential Privacy in Forward Pass
• Synthetic Data: Revisiting the Privacy-Utility Trade-off
• Membership Inference Attacks and Generalization: A Causal Perspective
• NeuGuard: Lightweight Neuron-Guided Defense against Membership Inference Attacks
• SLMIA-SR: Speaker-Level Membership Inference Attacks against Speaker Recognition Systems
• MUSO: Achieving Exact Machine Unlearning in Over-Parameterized Regimes
• Inferring Communities of Interest in Collaborative Learning-based Recommender Systems