Ant Group Research Institute Presents Two First-Author Papers at USENIX Security 2023 on Secure MPC for GBDT Training and Efficient 3PC for Binary Circuits

The 32nd USENIX Security Symposium was held from August 9‑11 in Anaheim, California, and is recognized as one of the four top international conferences in computer security alongside IEEE S&P, ACM CCS, and NDSS.

Ant Group’s Research Institute acted as a sponsor for the conference, supporting its activities and promoting its research.

Two first‑author papers from Ant Group’s Cryptography Laboratory were accepted. The first, titled “Squirrel: A Scalable Secure Two‑Party Computation Framework for Training Gradient Boosting Decision Trees,” presents a framework that enables secure, high‑performance training of GBDT models using MPC, achieving more than 30× speedup over the previous best system (Pivot, VLDB 2020) and comparable speed to SecureBoost while offering stronger security.

The second paper, “Efficient 3PC for Binary Circuits with Application to Maliciously‑Secure DNN Inference,” co‑authored with Tsinghua University and Alibaba Group, introduces a more efficient three‑party computation protocol for binary circuits, which are essential for evaluating non‑linear functions in MPC (e.g., ReLU, Gelu, softmax). The approach leverages distributed zero‑knowledge proofs to verify the correctness of the three‑party Boolean computation.

Ant Group’s Research Institute continues to engage with the academic community, contributing to leading security conferences and advancing core technologies for the digital and intelligent future.