FheFL: Fully Homomorphic Encryption Friendly Privacy-Preserving Federated Learning with Byzantine Users¶
Authors: Yogachandran Rahulamathavan; Charuka Herath; Xiaolan Liu; Sangarapillai Lambotharan; Carsten Maple
Year: 2023
Venue: arXiv preprint
type: technique collection: ppml status: auto
Links: arXiv:2306.05112
Summary¶
Auto-extracted summary (unreviewed)
FheFL is a federated learning algorithm in which users' model updates are protected by a distributed multi-key additive homomorphic encryption scheme so the aggregation server never sees plaintext gradients. The authors also design an aggregation rule that operates in the encrypted domain and weights users by their non-poisoning rates, so that data-poisoning (Byzantine) users can be countered even though the updates are encrypted. Security, privacy and convergence analyses plus experiments are reported, with the claim that FheFL reaches accuracy comparable to standard FL at reasonable computational cost.
Abstract¶
The federated learning (FL) technique was developed to mitigate data privacy issues in the traditional machine learning paradigm. While FL ensures that a user's data always remain with the user, the gradients are shared with the centralized server to build the global model. This results in privacy leakage, where the server can infer private information from the shared gradients. To mitigate this flaw, the next-generation FL architectures proposed encryption and anonymization techniques to protect the model updates from the server. However, this approach creates other challenges, such as malicious users sharing false gradients. Since the gradients are encrypted, the server is unable to identify rogue users. To mitigate both attacks, this paper proposes a novel FL algorithm based on a fully homomorphic encryption (FHE) scheme. We develop a distributed multi-key additive homomorphic encryption scheme that supports model aggregation in FL. We also develop a novel aggregation scheme within the encrypted domain, utilizing users' non-poisoning rates, to effectively address data poisoning attacks while ensuring privacy is preserved by the proposed encryption scheme. Rigorous security, privacy, convergence, and experimental analyses have been provided to show that FheFL is novel, secure, and private, and achieves comparable accuracy at reasonable computational cost.
Source: arxiv
Extracted details¶
| Field | Value | Source | Confidence |
|---|---|---|---|
| contribution | Distributed multi-key additive homomorphic encryption scheme for FL model aggregation | llm | 0.60 |
| contribution | Encrypted-domain aggregation using users' non-poisoning rates to mitigate data poisoning attacks | llm | 0.60 |
| contribution | Security, privacy and convergence analyses of the resulting FL protocol | llm | 0.60 |
| contribution | Experimental evaluation showing comparable accuracy at reasonable cost | llm | 0.60 |
| schemes | CKKS | llm | 0.60 |
| schemes | multi-key additive homomorphic encryption | llm | 0.60 |
| tasks | federated learning | llm | 0.60 |
| tasks | Byzantine-robust aggregation | llm | 0.60 |
| tasks | defense against data poisoning | llm | 0.60 |
| tasks | privacy-preserving model aggregation | llm | 0.60 |
| security level | 128-bit (ring dimension N between 8192 and 16384) | llm | 0.60 |
| threat model | honest-but-curious aggregation server inferring from gradients; malicious (Byzantine) users submitting poisoned updates | llm | 0.60 |
| interaction | other | llm | 0.60 |
| llm relevance | low | llm | 0.60 |
Keyword mentions¶
Counts of vocabulary terms in the full text (first page of occurrence in brackets).
- scheme: CKKS 25 [p.1], Secret sharing 7 [p.3], Paillier 6 [p.3], Multi-key HE 5 [p.6], Differential privacy 3 [p.1]
- library: TenSEAL 2 [p.10], PyTorch 1 [p.10]
- model: CNN 4 [p.10], ResNet 1 [p.11], Logistic regression 1 [p.13]
- dataset: CIFAR-100 21 [p.3], MNIST 17 [p.10], CIFAR-10 17 [p.10]
- technique: Federated learning 21 [p.1], Secure aggregation 7 [p.2], SIMD packing 5 [p.4], Two-party computation 5 [p.3], Convolution 3 [p.10], Malicious model 3 [p.1], RNS 2 [p.6], Quantization 2 [p.4], Membership inference 2 [p.2]
- assumption: LWE 9 [p.5], Lattice-based 5 [p.5], RLWE 4 [p.5]
- topic: Homomorphic encryption 59 [p.1]
Files¶
| Path (under papers/) | Collection | Pages | Size | SHA-256 | Note |
|---|---|---|---|---|---|
references/ppml/federated-learning/rahulamathavan-2023_fhefl-fhe-friendly-privacy-preserving-federated-learning.pdf |
ppml/federated-learning | 16 | 16.3 MB | f422cb33cec4 |
was Rahulamathavan et al. - 2023 - FheFL Fully Homomorphic Encryption Friendly Privacy-Preserving Federated Learning with Byzantine Us.pdf |
Database key: fhefl-fully-homomorphic-encryption-friendly-privacy-preserving · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set fhefl-fully-homomorphic-encryption-friendly-privacy-preserving COLUMN VALUE