Fully homomophic encryption over the integers revisited

Abstract

Two main computational problems serve as security foundations of current fully homomorphic encryption schemes: Regevs Learning With Errors problem (LWE) and Howgrave-Grahams Approximate Greatest Common Divisor problem (AGCD). Our first contribution is a reduction from LWE to AGCD. As a second contribution, we describe a new AGCD-based fully homomorphic encryption scheme, which outperforms all prior AGCD-based proposals: its security does not rely on the presumed hardness of the so-called Sparse Subset Sum problem, and the bit-length of a ciphertext is only Õ(λ), where λ refers to the security parameter. © International Association for Cryptologic Research 2015.