Skip to content

Federated Learning: An approach with Hybrid Homomorphic Encryption

Authors: Pedro Correia; Ivan Silva; Ivone Amorim; Eva Maia; Isabel Praça
Year: 2025
Venue: arXiv preprint

type: system collection: ppml status: auto

Links: arXiv:2509.03427

Summary

Auto-extracted summary (unreviewed)

The paper presents the first hybrid homomorphic encryption framework for federated learning: clients encrypt model updates with the HE-friendly PASTA symmetric cipher and send the BFV-encrypted PASTA key, and the server homomorphically evaluates PASTA decryption (transciphering) before aggregating BFV ciphertexts. A Flower-based prototype on IID MNIST with 12 clients and 10 rounds reaches 97.6% accuracy (1.3% below plaintext), cuts client upload bandwidth by over 2,000x and client runtime by 30% relative to BFV-only FL, but server computation grows by roughly 15,621x per client.

Abstract

Federated Learning (FL) is a distributed machine learning approach that promises privacy by keeping the data on the device. However, gradient reconstruction and membership-inference attacks show that model updates still leak information. Fully Homomorphic Encryption (FHE) can address those privacy concerns but it suffers from ciphertext expansion and requires prohibitive overhead on resource-constrained devices. We propose the first Hybrid Homomorphic Encryption (HHE) framework for FL that pairs the PASTA symmetric cipher with the BFV FHE scheme. Clients encrypt local model updates with PASTA and send both the lightweight ciphertexts and the PASTA key (itself BFV-encrypted) to the server, which performs a homomorphic evaluation of the decryption circuit of PASTA and aggregates the resulting BFV ciphertexts. A prototype implementation, developed on top of the Flower FL framework, shows that on independently and identically distributed MNIST dataset with 12 clients and 10 training rounds, the proposed HHE system achieves 97.6% accuracy, just 1.3% below plaintext, while reducing client upload bandwidth by over 2,000x and cutting client runtime by 30% compared to a system based solely on the BFV FHE scheme. However, server computational cost increases by roughly 15621x for each client participating in the training phase, a challenge to be addressed in future work.
Source: arxiv

Extracted details

Field Value Source Confidence
contribution first hybrid homomorphic encryption (PASTA + BFV) framework for federated learning llm 0.85
contribution transciphering-based secure aggregation that reduces client bandwidth and computation llm 0.85
contribution prototype on the Flower FL framework evaluated on MNIST llm 0.85
schemes BFV llm 0.85
schemes PASTA llm 0.85
schemes hybrid homomorphic encryption (transciphering) llm 0.85
libraries Flower llm 0.85
libraries hybrid-HE-framework (PASTA authors' C++ framework) llm 0.85
libraries pybind11 llm 0.85
datasets MNIST llm 0.85
tasks federated learning llm 0.85
tasks secure model aggregation llm 0.85
tasks image classification llm 0.85
hardware Google Colab llm 0.85
threat model honest-but-curious server, honest clients, no server-client collusion, trusted setup by a third-party authority llm 0.85
interaction non-interactive FHE llm 0.85
code url https://github.com/isec-tugraz/hybrid-HE-framework (URL found in the full text) regex 0.50
code url https://github.com/tfhe/tfhe (URL found in the full text) regex 0.50
code url https://github.com/pybind/pybind11 (URL found in the full text) regex 0.50
limitations server-side computation increases by about 15,621x per client llm 0.85
limitations BFV lacks real-number division, so clients perform weighted averaging locally (potential privacy issue) llm 0.85
limitations single shared HE key offers no protection against server-client collusion llm 0.85
limitations no dropout-resilience mechanism llm 0.85
llm relevance low llm 0.85
key result accuracy: 97.6% (1.3% below plaintext) (IID MNIST, 12 clients, 10 rounds) llm 0.85
key result client upload bandwidth reduction: over 2,000x (vs BFV-only FL) llm 0.85
key result client runtime reduction: 30% (vs BFV-only FL) llm 0.85
key result client encryption/decryption speed-up: almost 10x (vs BFV-only FL) llm 0.85
key result server computation overhead: about 15,621x per client (homomorphic evaluation of PASTA decryption, Google Colab) llm 0.85

Keyword mentions

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

  • scheme: BFV 49 [p.1], Hybrid HE / transciphering 46 [p.1], CKKS 9 [p.4], BGV 7 [p.4], TFHE 5 [p.4], MPC 5 [p.2], Secret sharing 3 [p.6], Partially homomorphic encryption 3 [p.4], Multi-key HE 2 [p.18], Leveled / somewhat HE 2 [p.4]
  • library: Microsoft SEAL 2 [p.6], HElib 2 [p.6], TenSEAL 2 [p.13]
  • model: CNN 4 [p.13]
  • dataset: MNIST 4 [p.1]
  • hardware: NVIDIA T4 2 [p.13], GPU 2 [p.13], Intel Xeon 1 [p.15]
  • technique: Federated learning 32 [p.1], Quantization 22 [p.12], Malicious model 3 [p.4], Membership inference 3 [p.2], Convolution 2 [p.13], Secure aggregation 2 [p.7], Semi-honest model 2 [p.4]
  • task: IoT / edge 9 [p.1], Medical / healthcare 2 [p.1]
  • topic: Homomorphic encryption 85 [p.1], Open source 6 [p.6], Benchmark 2 [p.6]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/federated-learning/correia-2025_federated-learning-an-approach-with-hybrid-homomorphic-encryption.pdf ppml/federated-learning 19 1.2 MB 75c029bb7d0a was Correia et al. - 2025 - Federated Learning An approach with Hybrid Homomorphic Encryption.pdf

Database key: federated-learning-an-approach-with-hybrid-homomorphic-encryption · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set federated-learning-an-approach-with-hybrid-homomorphic-encryption COLUMN VALUE