论文笔记：Secure Federated Matrix Factorization_federated matrix factorization with privacy guaran-CSDN博客

本文链接：https://blog.csdn.net/ChangeOrange/article/details/114294876

FedMF是一种隐私保护的矩阵分解方法，用于推荐系统，防止用户偏好数据和潜在特征向量泄露。通过使用加性同态加密，FedMF在用户级别分布式地执行矩阵分解，确保在诚实但好奇的服务器上进行安全的协同过滤。实验证明，FedMF能够保护用户隐私，同时在小规模物品集上保持可接受的计算效率。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

1. Introduction

In recommendation systems, two types of users’ private information are leaked in traditional MF（Matrix Factorization） ：
(i) users’ raw preference data
(ii)users’ learned latent feature vectors

Prior studies have studied privacy-preserving MF in two main types：
(1) Obfuscation-based methods：obfuscate users’ preference raw data before releasing it to the central server so as to ensure certain level of privacy protection (e.g., differential privacy)
(2) Encryption-based methods：use advanced encryption schemes such as homomorphic encryption for implementing privacy-preserving MF

2.Preliminaries

Horizontal Federated Learning

Horizontal federated learning is introduced in the scenarios when the data from different contributors share the same feature space but vary in samples.

Additively Homomorphic Encryption
在这里插入图片描述

3.User-level Distributed Matrix Factorization

M ∈ [n] × [m] as user-item rating pairs which a rating has been generated, M = |M| as the total number of ratings and r_i,j represents the rating generated by user i for item j

Matrix factorization formulates this problem as fitting a bi-linear model on the existing ratings. The computing process of U and V can be done by sol