A Verifiable Framework using Trusted Execution Environments for Privacy-Preserving Machine Learning

Machine learning is now widespread and utilized across various industries for mission-critical applications such as fraud detection, algorithmic trading, and predicting disease-spreading patterns. Machine learning models require rich data for training, processing input, and inference. Importantly, all these services are now deployed in third-party public clouds, where consumers have little choice but to place implicit trust in the cloud provider to protect the confidentiality of their data without any validation. This inevitably poses security and privacy challenges, including issues related to various regulatory and legal requirements such as the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). To address this, we propose a verifiable PPML framework using TEE that will address various privacy threats to Machine Learning as a Service (MLaaS). Our framework enables MLaaS providers to deliver their use cases seamlessly without requiring changes to their Machine Learning (ML) model or training data while providing verifiable trust to enterprises. We demonstrate the framework through a public cloud proof of concept (POC).