Paper 2015/692

Fast and Secure Linear Regression and Biometric Authentication with Security Update

Yoshinori Aono, Takuya Hayashi, Le Trieu Phong, and Lihua Wang

Abstract

We explicitly present a homomorphic encryption scheme with a flexible encoding of plaintexts. We prove its security under the LWE assumption, and innovatively show how the scheme can be used to handle computations over both binary strings and real numbers. In addition, using the scheme and its features, we build fast and secure systems of - linear regression using gradient descent, namely finding a reasonable linear relation between data items which remain encrypted. Compared to the best previous work over a simulated dataset of $10^8$ records each with 20 features, our system dramatically reduces the server running time from about 8.75 hours (of the previous work) to only about 10 minutes. - biometric authentication, in which we show how to reduce ciphertext sizes by half and to do the computation at the server very fast, compared with the state-of-the-art. Moreover, as key rotation is a vital task in practice and is recommended by many authorized organizations for key management, - we show how to do key rotation over encrypted data, without any decryption involved, and yet homomorphic properties of ciphertexts remain unchanged. In addition, our method of doing key rotation handles keys of different security levels (e.g., 80- and 128-bit securities), so that the security of ciphertexts and keys in our scheme can be "updated", namely can be changed into a higher security level.

Note: Add publication information.