Skip to content

An Efficient and Privacy-Preserving Federated Learning Approach Based on Homomorphic Encryption

Authors: Francesco Castro; Donato Impedovo; Giuseppe Pirlo
Year: 2025
Venue: IEEE Open Journal of the Computer Society

type: technique collection: ppml status: auto

Links: DOI

Summary

Auto-extracted summary (unreviewed)

The paper proposes an efficient privacy-preserving federated learning method in which local model weights are quantized with a dynamic range evaluation layer-for-layer (DREL) scheme into long signed integers and encrypted with the integer-friendly BFV scheme at a 192-bit security level, reducing quantization error, communication and cryptographic cost. Across five datasets and four models (three CNNs and a feedforward network) the method needs fewer FL rounds to converge and, compared with Paillier- and CKKS-based PPFL-HE, cuts encryption time by 99.95% and 73.79%, decryption time by 99.90% and 55.13%, and ciphertext size by 5.78% and 75.17% respectively.

Abstract

Federated Learning (FL) is a decentralized and collaborative learning approach that ensures the data privacy of each participant. However, recent studies have shown that the private data of each participant can be obtained from shared parameters of local models through reversal model and membership inference attacks leading to privacy leakage. Privacy-preserving federated learning strategies based on Homomorphic Encryption (PPFL-HE) have been developed to solve this issue. PPFL-HE methods require high communication and computational overheads, which are impractical for resource-limited devices. This work proposes an efficient PPFL-HE method to reduce communication and computational overheads. The proposed method is based on an innovative quantization process that introduces a dynamic range evaluation layer-for-layer (DREL) to encode the weights of the local models into long-signed integers. Compared to standard quantization approaches, the proposed method reduces the quantization errors and the communication overhead. Moreover, it enables the encryption of local weights with the Brakerski/Fan-Vercauteren Homomorphic Encryption scheme (BFV-HE), which is highly efficient on integers, reducing encryption, aggregation, decryption time, and ciphertext size. The experiments conducted with five popular datasets and four different Machine Learning (ML) models (three CNN models and a feedforward neural network) show that the proposed method is more efficient in communication and computational overheads than other PPFL-HE methods. Specifically, the proposed method requires fewer FL rounds to achieve global model convergence and leads to an average reduction in encryption time of 99.95% and 73.79%, in decryption time of 99.90% and 55.13%, and in ciphertext size of 5.78% and 75.17% compared to PPFL-HE methods based on Paillier and CKKS schemes, respectively.
Source: pdf

Extracted details

Field Value Source Confidence
contribution DREL quantization (dynamic range evaluation layer-for-layer) encoding weights as long signed integers llm 0.75
contribution BFV-based encrypted aggregation of local model weights in federated learning llm 0.75
contribution reduced quantization error and communication overhead vs standard quantization llm 0.75
contribution comparison against Paillier- and CKKS-based PPFL-HE methods on five datasets and four models llm 0.75
schemes BFV llm 0.75
schemes Paillier llm 0.75
schemes CKKS llm 0.75
models CNN (three models) llm 0.75
models feedforward neural network llm 0.75
tasks federated learning llm 0.75
tasks privacy-preserving model aggregation llm 0.75
security level 192-bit llm 0.75
threat model untrusted aggregation server (model inversion / membership inference on shared weights) llm 0.75
interaction other llm 0.75
llm relevance low llm 0.75
key result encryption time reduction: 99.95% (vs Paillier-based PPFL-HE, average) llm 0.75
key result decryption time reduction: 99.90% (vs Paillier-based PPFL-HE, average) llm 0.75
key result ciphertext size reduction: 5.78% (vs Paillier-based PPFL-HE, average) llm 0.75
key result encryption time reduction: 73.79% (vs CKKS-based PPFL-HE, average) llm 0.75
key result decryption time reduction: 55.13% (vs CKKS-based PPFL-HE, average) llm 0.75
key result ciphertext size reduction: 75.17% (vs CKKS-based PPFL-HE, average) llm 0.75

Keyword mentions

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

  • scheme: BFV 50 [p.1], CKKS 21 [p.1], Paillier 19 [p.1], ElGamal 3 [p.2], BGV 2 [p.2], Differential privacy 2 [p.2]
  • model: CNN 7 [p.1], Logistic regression 1 [p.11]
  • dataset: MNIST 5 [p.6], SVHN 3 [p.6]
  • hardware: GPU 1 [p.7]
  • technique: Quantization 90 [p.1], Federated learning 28 [p.1], Secure aggregation 4 [p.6]
  • assumption: RLWE 7 [p.3], IND-CPA 5 [p.10], LWE 2 [p.3]
  • task: IoT / edge 8 [p.1], Medical / healthcare 7 [p.1]
  • topic: Homomorphic encryption 75 [p.1]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/federated-learning/castro-2025_efficient-and-privacy-preserving-federated-learning-based-on-he.pdf ppml/federated-learning 12 2.3 MB 665892f7faee was Castro et al. - 2025 - An Efficient and Privacy-Preserving Federated Learning Approach Based on Homomorphic Encryption.pdf

Database key: an-efficient-and-privacy-preserving-federated-learning-approach-based · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set an-efficient-and-privacy-preserving-federated-learning-approach-based COLUMN VALUE