Secure and efficient general matrix multiplication on cloud using homomorphic encryption

Abstract

Matrix multiplication is a common operation in scientific computing, but can be com- putationally expensive. Cloud computing provides a scalable solution, but using it for sensitive data, such as medical records, raises privacy concerns. Homomorphic encryp- tion allows cloud computation service providers to operate on the encrypted matrices without decryption, but its high computational cost makes it impractical for real-world large-scale tasks. A previous study by Yang Gao et al. introduced the HEGMM (Ho- momorphic Encryption General Matrix Multiplication) algorithm, which uses SIMD (Single instruction, Multiple Data) operations to overcome this bottleneck. In this project, we build on their work by implementing 2 matrix packing methods to further improve the performance of HEGMM.