Privacy-Aware Quadratic Optimization Using Partially Homomorphic Encryption

Yasser Shoukry, Konstantinos Gatsis, Amr Alanwar, George J. Pappas, Sanjit A. Seshia andMani Srivastava, and Paulo Tabuada. Privacy-Aware Quadratic Optimization Using Partially Homomorphic Encryption. In Proceedings of the 55th IEEE Conference on Decision and Control (CDC), pp. 5053–5058, December 2016.

Download

[pdf] 

Abstract

We consider the problem where multiple agents are interested in solving a joint optimization problem in a privacy-preserving manner. In this setup, all the agents jointly optimize a quadratic objective function subject to linear inequality constraints where the parameters of the objective function as well as the constraints are privacy-sensitive and possessed by different agents. We propose a privacy-preserving protocol based on partially homomorphic encryption where each agent encrypts their own information before sending it to an untrusted cloud computing infrastructure. The cloud applies gradient descent type algorithms on the encrypted data without the ability to decrypt them. The privacy of the proposed protocol, against coalitions of colluding agents, is then analyzed using the cryptography notion of zero knowledge proofs.

BibTeX

@inproceedings{shoukry-cdc16b,
  author    = {Yasser Shoukry and Konstantinos Gatsis and Amr Alanwar and George J. Pappas and Sanjit A. Seshia and
Mani Srivastava and Paulo Tabuada},
  title     = {Privacy-Aware Quadratic Optimization Using Partially Homomorphic Encryption},
  booktitle = {Proceedings of the 55th IEEE Conference on Decision and Control (CDC)},
  Year = {2016},
  Month = {December},
  pages = "5053--5058",
  abstract = {We consider the problem where multiple agents are interested in solving a joint optimization problem in a privacy-preserving manner. In this setup, all the agents jointly optimize a quadratic objective function subject to linear inequality constraints where the parameters of the objective function as well as the constraints are privacy-sensitive and possessed by different agents. We propose a privacy-preserving protocol based on partially homomorphic encryption where each agent encrypts their own information before sending it to an  untrusted cloud computing infrastructure. The cloud applies gradient descent type algorithms on the encrypted data without the ability to decrypt them. The privacy of the proposed protocol, against coalitions of colluding agents, is then analyzed using the cryptography notion of zero knowledge proofs.},
}