Evaluation of Computation Overhead of Paillier Encryption in Vertical Federated Learning
P Abhinand, Tomsy Paul, G. Santhosh Kumar · 2024
Paillier Encryption is one of the most commonly used Partially Homomorphic Encryption (PHE) techniques. Since it is additive and also faster compared with fully homomorphic techniques, it is used extensively in Vertical Federated Learning (VFL) where encryption is required to secure the exchange of gradients and partial results between parties. However, the computation associated with Paillier encryption is very huge. We present its computation overhead in the context of VFL for various machine learning models. Specifically, we find the number of encryption/decryption and homomorphic operations such as addition and scalar multiplication required in various machine learning models and the time of execution associated with each. We see that as the number of iterations increases, these operations have a severe impact on the total time for learning.