Rhombus: Fast Homomorphic Matrix‑Vector Multiplication for Secure Two‑Party Inference – Paper Overview and Live Presentation

With the rapid development of big data and artificial intelligence, privacy‑preserving machine learning (PPML) has become a hot topic in both academia and industry. PPML aims to enable collaborative model training and inference without exposing raw data, relying heavily on homomorphic encryption (HE) and secure multi‑party computation (SMPC).

The paper "Rhombus: Fast Homomorphic Matrix‑Vector Multiplication for Secure Two‑Party Inference" proposes a new protocol that significantly improves the performance of matrix‑vector multiplication (MVM) in PPML. By reducing the number of ciphertext rotations and maintaining O(1) communication complexity, Rhombus offers a more efficient solution compared to prior HE‑based methods.

In the two‑party privacy inference scenario, Rhombus achieves lower computational cost and communication overhead, making HE‑based MVM more practical for real‑world applications such as healthcare, finance, and education.

The article also announces a live streaming event jointly organized by Ant Technology Research Institute and the SecretFlow community, where Ant Digital Technology senior algorithm engineer and first author He Jiaxing will present an in‑depth walkthrough of the Rhombus protocol, covering its technical details and experimental evaluation.

Viewers can watch the live stream on WeChat Video Channels (Ant Technology Research Institute, Ant Technology), SecretFlow, and Bilibili on November 14, 2024, from 19:00 to 20:00.