Optimizing Homomorphic Encryption in Federated Learning with Zero-Skipping

View at publisher Summarise with VIRAT
Authors: Yoo-Bin Tae, Soo-Jeong Park, Geonha Kim, Seung‐Ho Lim
Cited by: 0
References: 7

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Federated learning enables training AI model without sharing of data, but requires protections such as homomorphic encryption to prevent reverse engineering of shared gradients, which incurs significant computational overhead. To mitigate this overhead, we propose an improved compression scheme for homomorphic encryption that integrates two key optimizations: clipped quantization and zero-skipping encryption. Clipped quantization increases the proportion of zeros during gradient quantization, while zero-skipping encryption identifies blocks that are quantized to zero and omits homomorphic encryption, thereby reducing the overhead of homomorphic encryption. The preliminary experiments for the federated learning with CNN show that the improved homomorphic encryption method could reduce client-side homomorphic encryption time and overall training time with a small decrease in accuracy.

Fields of Study & Subjects

Privacy-Preserving Technologies in Data Cryptography and Data Security Cloud Data Security Solutions

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–2
Publication date
3 February 2026
Language
EN