Skip to content

HHEML: Hybrid Homomorphic Encryption for Privacy-Preserving Machine Learning on Edge

Authors: Yu Hin Chan; Hao Yang; Shiyu Shen; Xingyu Fan; Shengzhe Lyu; Patrick S. Y. Hung; Ray C. C. Cheung
Year: 2025
Venue: arXiv preprint

type: accelerator collection: ppml status: auto

Links: arXiv:2510.20243

Summary

Auto-extracted summary (unreviewed)

HHEML proposes a hardware-accelerated hybrid homomorphic encryption (HHE) architecture for privacy-preserving ML inference on edge devices, built around the FHE-friendly Pasta symmetric cipher implemented as a dedicated FPGA accelerator with several microarchitectural optimizations and integrated into an end-to-end PPML pipeline. On a PYNQ-Z2 board with MNIST, the authors report over a 50x reduction in client-side encryption latency and nearly 2x higher hardware throughput than existing FPGA-based HHE accelerators.

Abstract

Privacy-preserving machine learning (PPML) is an emerging topic to handle secure machine learning inference over sensitive data in untrusted environments. Fully homomorphic encryption (FHE) enables computation directly on encrypted data on the server side, making it a promising approach for PPML. However, it introduces significant communication and computation overhead on the client side, making it impractical for edge devices. Hybrid homomorphic encryption (HHE) addresses this limitation by combining symmetric encryption (SE) with FHE to reduce the computational cost on the client side, and combining with an FHE-friendly SE can also lessen the processing overhead on the server side, making it a more balanced and efficient alternative. Our work proposes a hardware-accelerated HHE architecture built around a lightweight symmetric cipher optimized for FHE compatibility and implemented as a dedicated hardware accelerator. To the best of our knowledge, this is the first design to integrate an end-to-end HHE framework with hardware acceleration. Beyond this, we also present several microarchitectural optimizations to achieve higher performance and energy efficiency. The proposed work is integrated into a full PPML pipeline, enabling secure inference with significantly lower latency and power consumption than software implementations. Our contributions validate the feasibility of low-power, hardware- accelerated HHE for edge deployment and provide a hardware- software co-design methodology for building scalable, secure machine learning systems in resource-constrained environments. Experiments on a PYNQ-Z2 platform with the MNIST dataset show over a 50x reduction in client-side encryption latency and nearly a 2x gain in hardware throughput compared to existing FPGA-based HHE accelerators.
Source: arxiv

Extracted details

Field Value Source Confidence
contribution first end-to-end HHE framework integrated with hardware acceleration llm 0.70
contribution FPGA accelerator for the FHE-friendly Pasta symmetric cipher llm 0.70
contribution microarchitectural optimizations for higher performance and energy efficiency llm 0.70
contribution integration into a full PPML inference pipeline for edge devices llm 0.70
contribution hardware-software co-design methodology for resource-constrained secure ML llm 0.70
schemes hybrid homomorphic encryption (HHE) llm 0.70
schemes Pasta symmetric cipher llm 0.70
schemes FHE llm 0.70
datasets MNIST llm 0.70
tasks privacy-preserving ML inference on edge/IoT devices llm 0.70
tasks image classification llm 0.70
hardware PYNQ-Z2 FPGA llm 0.70
threat model untrusted computation server (client-server) llm 0.70
interaction non-interactive FHE llm 0.70
llm relevance low llm 0.70
key result client-side encryption latency reduction: >50x (PYNQ-Z2, MNIST) llm 0.70
key result hardware throughput gain: ~2x (vs existing FPGA-based HHE accelerators) llm 0.70

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • scheme: Hybrid HE / transciphering 49 [p.1], MPC 4 [p.1], CKKS 2 [p.1], BFV 2 [p.1]
  • library: Phantom 2 [p.2], Cheddar 2 [p.2]
  • model: ResNet 1 [p.3]
  • dataset: MNIST 10 [p.1]
  • hardware: FPGA 22 [p.1], ASIC 5 [p.2], ARM 2 [p.7], GPU 1 [p.2]
  • technique: Bootstrapping 2 [p.2], Polynomial approximation 2 [p.3]
  • task: IoT / edge 7 [p.1], Finance 2 [p.1]
  • topic: Homomorphic encryption 89 [p.1], Privacy-preserving machine learning 37 [p.1], Private / secure inference 4 [p.1], Benchmark 3 [p.2]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/inference/chan-2025_hheml-hybrid-homomorphic-encryption-for-ppml-on-edge.pdf ppml/inference 9 450.5 KB 23b775213e56 was Chan et al. - 2025 - HHEML Hybrid Homomorphic Encryption for Privacy-Preserving Machine Learning on Edge.pdf

Database key: hheml-hybrid-homomorphic-encryption-for-privacy-preserving-machine · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set hheml-hybrid-homomorphic-encryption-for-privacy-preserving-machine COLUMN VALUE