Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GuardML: Efficient Privacy-Preserving Machine Learning Services Through Hybrid Homomorphic Encryption

Eugene Frimpong, Khoa Nguyen, Mindaugas Budzys, Tanveer Khan, Antonis Michalas

TL;DR

GuardML leverages Hybrid Homomorphic Encryption to enable privacy-preserving inference on end devices, addressing the efficiency barrier of traditional HE in PPML. By combining symmetric-key encryption with HE (PASTA-based HHE), the framework offloads heavy computations to a CSP while preserving input and model privacy, validated on an ECG-based heart disease classification task. The two protocols, 2GML and 3GML, cover both 2-party and 3-party settings with formal security analyses and practical evaluations showing near-plaintext accuracy and favorable communication/computation trade-offs. This work demonstrates a viable path toward practical, secure MLaaS-like services on constrained devices, with measurable gains in efficiency and strong privacy guarantees in real-world deployments.

Abstract

Machine Learning (ML) has emerged as one of data science's most transformative and influential domains. However, the widespread adoption of ML introduces privacy-related concerns owing to the increasing number of malicious attacks targeting ML models. To address these concerns, Privacy-Preserving Machine Learning (PPML) methods have been introduced to safeguard the privacy and security of ML models. One such approach is the use of Homomorphic Encryption (HE). However, the significant drawbacks and inefficiencies of traditional HE render it impractical for highly scalable scenarios. Fortunately, a modern cryptographic scheme, Hybrid Homomorphic Encryption (HHE), has recently emerged, combining the strengths of symmetric cryptography and HE to surmount these challenges. Our work seeks to introduce HHE to ML by designing a PPML scheme tailored for end devices. We leverage HHE as the fundamental building block to enable secure learning of classification outcomes over encrypted data, all while preserving the privacy of the input data and ML model. We demonstrate the real-world applicability of our construction by developing and evaluating an HHE-based PPML application for classifying heart disease based on sensitive ECG data. Notably, our evaluations revealed a slight reduction in accuracy compared to inference on plaintext data. Additionally, both the analyst and end devices experience minimal communication and computation costs, underscoring the practical viability of our approach. The successful integration of HHE into PPML provides a glimpse into a more secure and privacy-conscious future for machine learning on relatively constrained end devices.

GuardML: Efficient Privacy-Preserving Machine Learning Services Through Hybrid Homomorphic Encryption

TL;DR

GuardML leverages Hybrid Homomorphic Encryption to enable privacy-preserving inference on end devices, addressing the efficiency barrier of traditional HE in PPML. By combining symmetric-key encryption with HE (PASTA-based HHE), the framework offloads heavy computations to a CSP while preserving input and model privacy, validated on an ECG-based heart disease classification task. The two protocols, 2GML and 3GML, cover both 2-party and 3-party settings with formal security analyses and practical evaluations showing near-plaintext accuracy and favorable communication/computation trade-offs. This work demonstrates a viable path toward practical, secure MLaaS-like services on constrained devices, with measurable gains in efficiency and strong privacy guarantees in real-world deployments.

Abstract

Machine Learning (ML) has emerged as one of data science's most transformative and influential domains. However, the widespread adoption of ML introduces privacy-related concerns owing to the increasing number of malicious attacks targeting ML models. To address these concerns, Privacy-Preserving Machine Learning (PPML) methods have been introduced to safeguard the privacy and security of ML models. One such approach is the use of Homomorphic Encryption (HE). However, the significant drawbacks and inefficiencies of traditional HE render it impractical for highly scalable scenarios. Fortunately, a modern cryptographic scheme, Hybrid Homomorphic Encryption (HHE), has recently emerged, combining the strengths of symmetric cryptography and HE to surmount these challenges. Our work seeks to introduce HHE to ML by designing a PPML scheme tailored for end devices. We leverage HHE as the fundamental building block to enable secure learning of classification outcomes over encrypted data, all while preserving the privacy of the input data and ML model. We demonstrate the real-world applicability of our construction by developing and evaluating an HHE-based PPML application for classifying heart disease based on sensitive ECG data. Notably, our evaluations revealed a slight reduction in accuracy compared to inference on plaintext data. Additionally, both the analyst and end devices experience minimal communication and computation costs, underscoring the practical viability of our approach. The successful integration of HHE into PPML provides a glimpse into a more secure and privacy-conscious future for machine learning on relatively constrained end devices.
Paper Structure (18 sections, 2 theorems, 7 equations, 6 figures, 7 tables)

This paper contains 18 sections, 2 theorems, 7 equations, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Preliminaries
  4. Homomorphic Encryption
  5. Hybrid Homomorphic Encryption
  6. Machine Learning
  7. System Model
  8. GuardML
  9. GuardML: 2-Party Setting
  10. GuardML: 3-Party Setting
  11. Threat Model and Security Analysis
  12. Security Analysis
  13. Experiments Section
  14. PPML Application
  15. MIT-BIH dataset
  16. ...and 3 more sections

Key Result

proposition 1

Let $\sigma$ be an EUF-CMA secure signature scheme and $\mathsf{PKE}$ an INC-CPA public key encryption scheme. Then $\mathcal{ADV}$, cannot successfully launch the Ciphertext Substitution Attack against GuardML.

Figures (6)

  • Figure 1: $\mathsf{2GML}$
  • Figure 2: $\mathsf{3GML}$
  • Figure 3: Example heartbeats from the MIT-BIH dataset.
  • Figure 4: Quantizing ECG data into 4-bit integers. Top: Floating-point ECG data. Bottom: Corresponding quantized ECG data.
  • Figure 5: Computation Costs
  • ...and 1 more figures

Theorems & Definitions (6)

  • definition 1: Homomorphic Encryption
  • definition 2: Hybrid Homomorphic Encryption
  • proposition 1: Ciphertext Substitution Attack Soundness
  • proof
  • proposition 2: ML Model Unauthorized Access Attack Soundness
  • proof