GuardML: Efficient Privacy-Preserving Machine Learning – A Publication from Tampere University

4.3.2024 14:14

We’re sharing the latest publication by TUNI, a key partner in the Harpocrates project, focusing on privacy-preserving machine learning.

Big congrats to the authors: Eugene Frimpong, Khoa Nguyen, Mindaugas Budzys, Tanveer Khan, and Antonis Michalas!

Their paper, “GuardML: Efficient Privacy-Preserving Machine Learning Services Through Hybrid Homomorphic Encryption,” introduces a fresh approach to secure machine learning. GuardML uses Hybrid Homomorphic Encryption to keep input data and ML models private while securely learning classification outcomes over encrypted data.

Highlights:

• GuardML maintains data privacy with only a slight accuracy drop compared to plaintext data inference.
• It keeps communication and computation costs low for both analysts and end devices.

Check out the full publication here: https://arxiv.org/abs/2401.14840