Privacy-Aware Artificial Intelligence with Homomorphic Encryption using Machine Learning
Battula Srinivasa Rao, Saumitra Chattopadhyay, Prashant Singh, Bramah Hazela, G. Sabarinathan, Kalva Yamini · 2023
Along with the expansion of machine learning (ML) applications, the amount of data required to create predictions increases. Big-data ML has always been limited by off-chip memory capacity and computational speed. Considerably, privacy is one of the limitations of big data, which can be solved by homomorphic encryption (HE). Due to the combination of HE and ML, the multi-party privacy-protected ML suggested in this research may assist numerous users in doing artificial intelligence (AI) without disclosing private data. The technique may train common models in situations of data abuse, particularly in private data protection. The model trained using the ML technique named Artificial Neural Network (ANN) has a similar impact to the model developed using all data on a single computer, according to experiments using the algorithm. The gradient data is simply transmitted by all parties, and homomorphic procedures in the main computing system combine the gradient data. Besides, the optimal key is selected using the significance of the Lion Algorithm (LA). After homomorphic procedures, the learning model is modified depending on the new gradient data.