Coral: A Maliciously Secure Computation Framework for Packed and Mixed Circuits

On October 15, the three‑day ACM Computer and Communications Security (CCS) conference began in Salt Lake City, USA.

ACM CCS, together with IEEE S&P, USENIX Security, and NDSS, is regarded as one of the four top international academic conferences in information security and is also an A‑class conference recommended by the China Computer Federation (CCF).

Ant Group had eight papers accepted at ACM CCS 2024, with four of them having Ant Technology researchers as first authors.

Today we introduce one of these papers, independently completed by Ant Technology Research Institute’s Cryptography Lab: “Coral: Maliciously Secure Computation Framework for Packed and Mixed Circuits.”

The first author, researcher Zhichong Huang from Ant Technology Research Institute’s Cryptography Lab, presented the paper at ACM CCS 2024.

Coral is the lab’s latest research result and continues several representative works in secure computation.

Compared with the lab’s earlier general secure computation work, Coral’s main contribution is strengthening the system security model from semi‑honest to malicious, maintaining the intended security even when participants launch arbitrary malicious attacks, thereby significantly raising the system’s security level.

The paper proposes a more efficient malicious‑secure MPC framework that employs a mathematical tool called RMFE (reverse multiplication‑friendly embedding), essentially an encoding method. When applying the same computation circuit to multiple inputs (e.g., classifying several images simultaneously), RMFE allows a single error‑correcting code to verify all inputs, speeding up the protocol and reducing communication.

Earlier MPC work based on RMFE was purely theoretical and lacked implementations, making performance claims uncertain; in practice we found textbook implementations of RMFE problematic. Coral contributes the first practical, efficient implementation of RMFE, fixing several theoretical flaws of prior work, which is significant for future research.

The paper further integrates RMFE into a complete computation framework supporting both Boolean and arithmetic operations, and combined with the MPSPDZ framework, enables malicious‑model secure computation for general scenarios such as decision trees and neural network inference, as illustrated below.

Below are the eight papers from Ant Group that were accepted at ACM CCS 2024.