Privacy-preserving Machine Learning through Data Obfuscation

Abstract

As machine learning becomes a practice and commodity, numerous cloud-based services and frameworks are provided to help customers develop and deploy machine learning applications. While it is prevalent to outsource model training and serving tasks in the cloud, it is important to protect the privacy of sensitive samples in the training dataset and prevent information leakage to untrusted third parties. Past work have shown that a malicious machine learning service provider or end user can easily extract critical information about the training samples, from the model parameters or even just model outputs. In this paper, we propose a novel and generic methodology to preserve the privacy of training data in machine learning applications. Specifically we introduce an obfuscate function and apply it to the training data before feeding them to the model training task. This function adds random noise to existing samples, or augments the dataset with new samples. By doing so sensitive information about the properties of individual samples, or statistical properties of a group of samples, is hidden. Meanwhile the model trained from the obfuscated dataset can still achieve high accuracy. With this approach, the customers can safely disclose the data or models to third-party providers or end users without the need to worry about data privacy. Our experiments show that this approach can effective defeat four existing types of machine learning privacy attacks at negligible accuracy cost.

References in corpus (14)

• Deep Residual Learning for Image Recognition
• Random Erasing Data Augmentation
• Stealing Machine Learning Models via Prediction APIs
• EMNIST: an extension of MNIST to handwritten letters
• Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning
• Chiron: Privacy-preserving Machine Learning as a Service
• Understanding Membership Inferences on Well-Generalized Learning Models
• Stealing Hyperparameters in Machine Learning
• Towards Reverse-Engineering Black-Box Neural Networks
• Generating Artificial Data for Private Deep Learning
• Machine Learning Models that Remember Too Much
• Crowd-ML: A Privacy-Preserving Learning Framework for a Crowd of Smart Devices
• LOGAN: Membership Inference Attacks Against Generative Models
• Performing Co-Membership Attacks Against Deep Generative Models

Cited by in corpus (19)

• Privacy in Deep Learning: A Survey
• When Machine Learning Meets Privacy: A Survey and Outlook
• Privacy-Preserving Brain-Computer Interfaces: A Systematic Review
• Data Minimization for GDPR Compliance in Machine Learning Models
• Artificial Intelligence for Digital Agriculture at Scale: Techniques, Policies, and Challenges
• Anonymizing Machine Learning Models
• On analyzing and evaluating privacy measures for social networks under active attack
• AnonymousNet: Natural Face De-Identification with Measurable Privacy
• Game-Theoretic and Machine Learning-based Approaches for Defensive Deception: A Survey
• Towards Security Threats of Deep Learning Systems: A Survey
• Segmentations-Leak: Membership Inference Attacks and Defenses in Semantic Image Segmentation
• Flow-Loss: Learning Cardinality Estimates That Matter
• Key Protected Classification for Collaborative Learning
• Data Privacy Preservation on the Internet of Things
• The Value of Collaboration in Convex Machine Learning with Differential Privacy
• Privacy-Preserving Medical Image Classification through Deep Learning and Matrix Decomposition
• An Adaptive and Fast Convergent Approach to Differentially Private Deep Learning
• An Extension of Fano's Inequality for Characterizing Model Susceptibility to Membership Inference Attacks
• Reducing Risk of Model Inversion Using Privacy-Guided Training