An Adaptive User-Guided Resilient Approach for Dynamic Bootstrapping in Homomorphic Encryption

Cheon-Kim-Kim-Song (CKKS)-based Homomorphic Encryption (HE) allows encrypted data to undergo approximate calculations. This makes it particularly suitable for real-world applications that rely on floating-point operations, like signal processing and encrypted machine learning. Despite this advantage, most current systems use fixed bootstrapping schedules that activate regardless of the actual noise level in the encrypted data. This inflexible design leads to unnecessary bootstrapping, higher memory usage, and slower processing, especially when dealing with different data formats and file sizes. To overcome these challenges, we introduce the Adaptive User-guided Resilient Approach using CKKS (AURA-CKKS), a new encryption method featuring a dynamic, noise-sensitive bootstrapping process. In order to help the system decide whether bootstrapping is required, the AURA-CKKS algorithm first accepts user-defined parameters, such as noise thresholds and bootstrapping preferences. Before calculations start, the algorithm estimates noise growth by doing an initial examination of the ciphertext parameters. The algorithm constantly checks noise levels during encrypted operations to ensure that bootstrapping is only activated when required, improving efficiency and preventing unnecessary calculations. Throughout the homomorphic operation cycle, this adaptive technique preserves the integrity of the ciphertext, minimizes processing time, and permits effective management of computational resources. Test results show that AURA-CKKS can boost bootstrapping efficiency by up to 46%, reduce memory usage by around 39%, and increase processing speed by over 51% compared to standard CKKS methods. This positions AURA-CKKS as a powerful and adaptable solution for secure, encrypted computation. Experimental results demonstrate that AURA-CKKS significantly outperforms existing CKKS implementations in terms of throughput, scalability, and noise management, making it a practical and efficient solution for secure computation.