Learnable Encryption with a Diffusion Property
Ijaz Ahmad, Joongheon Kim, Seokjoo Shin · 2025
Guaranteeing privacy in outsourced deep learning (DL) model training ensures client control over their data and eases burden on service providers. Although learnable encryption can enable DL training in the encryption domain, its existing approaches do not adhere to cryptographic standards. Therefore, we propose a learnable encryption function that processes plaintext in a way that confusion and diffusion properties are ensured. Simulations on a COVID-19 dataset confirm its efficacy.