69、Learnable Encryption with Deniability: Protecting User Privacy in Cloud Machine Learning

Learnable Encryption with Deniability: Protecting User Privacy in Cloud Machine Learning

1. Introduction

Machine learning has become mainstream in recent years. With its diverse development, model training demands more resources, leading many to use cloud services for training, updating models, and making predictions. However, user privacy is a significant concern in cloud machine learning. When users upload training data or make prediction queries, the data is exposed to the cloud service provider (CSP), resulting in privacy leakage.

Existing solutions to this privacy issue include the integration of machine learning services with homomorphic encryption and learnable encryption. Homomorphic encryption simulates a prediction process as a circuit and ru