Our latest research results were accepted by ACM CCS2024

Recently, a research paper led by Associate Professor Peijia Zheng of our school was accepted by the 31st International Conference on Computer and Communications Security (ACM Conference on Computer and Communications Security, ACM CCS2024).

The title of the paper is "Two-Tier Data Packing in RLWE-based Homomorphic Encryption for Secure Federated Learning". The first author is Zhou Yufei, a 2022 master-doctoral student from the School of Computer Science. The corresponding author is Associate Professor Zheng Peijia from the School of Computer Science (Zhou Yufei's advisor). The paper was completed in collaboration with Professor Cao Xiaochun from the School of Cyberspace Security of Sun Yat-sen University and Professor Huang Jiwu from Shenzhen University of Science and Technology of China and Moscow State University. This is the first time that Sun Yat-sen University has published academic results at the ACM CCS conference as the first completing unit. Secure federated learning based on homomorphic encryption refers to the use of homomorphic encryption to protect the model parameters or gradient information that needs to be exchanged in the federated learning process, so as not to leak the privacy of the local data set during the training of federated learning. However, homomorphic encryption brings large communication and computational overheads, and how to improve the efficiency of computation and communication has always been an important challenge. The author proposed a novel data packing strategy TwoPack based on RLWE homomorphic encryption. Starting from the underlying algebraic structure, the author proposed a hierarchical double packaging strategy, namely the integer layer and the polynomial layer, to pack more data into the same ciphertext. Without changing the security level of homomorphic encryption, the parallelism of data is increased. On this basis, the author integrated the proposed TwoPack technology into the secure federated learning scheme under cross-silo and cross-device, and designed corresponding adaptation technology to improve the efficiency of existing secure aggregation. This data packaging technology is not only of great significance to secure federated learning, but also widely applicable to other privacy computing tasks.

It is reported that ACM CCS was first held in 1993 and has a history of more than 30 years. It is an internationally recognized top conference in the field of information security and a Class A conference recommended by the China Computer Federation (CCF). The acceptance rate of ACM CCS in the past ten years is about 18%, and the accepted manuscripts reflect the most cutting-edge research level in the field of information security. This ACM CCS will be held in Salt Lake City, USA from October 14 to 18, 2024.