Secure and efficient federated learning using attribute-based homomorphic encryption

doi:10.1016/j.jisa.2026.104564

Secure and efficient federated learning using attribute-based homomorphic encryption

2026-07-03

0

PRE

AI

I

Isaac Amankona Obiri *

E

Emmanuel Antwi‐Boasiako

E

Eric Kuada

A

Abigail Akosua Addobea

DOI：10.1016/j.jisa.2026.104564

Original

Original request for help

Save

Abstract

En 中文

• Proposes a Ciphertext-Policy Attribute-Based Homomorphic Encryption for Federated Learning (CP-ABHE-FL) framework, which embeds Linear Secret Sharing Scheme (LSSS)-based access policies into lattice-encrypted gradients, enabling policy-governed homomorphic aggregation without client interaction. • Establishes a mapping between Ciphertext-Policy Attribute-Based Encryption (CP-ABE) structures and lattice-based additive homomorphism, supporting expressive access policies including OR, AND, and threshold gates with non-interactive decryption independent of federation size. • Provides security guarantees, including indistinguishability under chosen-plaintext attack (IND-CPA), collusion resistance, and gradient integrity under the decisional ring learning with errors (DRLWE) assumption, while incorporating Gaussian differential privacy. • Achieves constant per-round communication overhead with respect to the number of clients, addressing the linear scalability limitations inherent in multi-key homomorphic encryption baselines.

Secure and efficient federated learning using attribute-based homomorphic encryption

Abstract

Journal

Organization