zoukankan      html  css  js  c++  java
  • Federated Recommondation

    Federated Recommondation

    1. Introduction

    Federatedrec aims to address recommondation problems such as rating prediction and item ranking under federated learning senario. It includes implementation of a number of popular recommondation algorithms based on FATE library. Such as federated fm, federated mf, federated svd etc.

    2. Background

    With laws related to the protection of data security and privacy, such as General Data Protection Regulation (GDPR), coming out in recent years, data collection becomes more difficult. And users give more attention to the problem of data privacy. Directly sharing user data between companies (organizaiotns) is undesired. Such data silo issues commonly exist in recommender systems.

    FedRec addresses the data silo issue and builds centralized recommender without compromising privacy and security. FedRecLib includes implementation of a suite of state-of-the-art recommondation algorithms based on FATE library.

    3. Algorithms list:

    1. Hetero FM(factorization machine)

    Build hetero factorization machine module through multiple parties.

    • Corresponding module name: HeteroFM
    • Data Input: Input DTable.
    • Model Output: Factorization Machine model.
    2. Homo-FM

    Build homo factorization machine module through multiple parties.

    • Corresponding module name: HomoFM
    • Data Input: Input DTable.
    • Model Output: Factorization Machine model.
    3. Hetero MF(matrix factorization)

    Build hetero matrix factorization module through multiple parties.

    • Corresponding module name: HeteroMF
    • Data Input: Input DTable of user-item rating matrix data.
    • Model Output: Matrix Factorization model.
    4. Hetero SVD

    Build hetero SVD module through multiple parties.

    • Corresponding module name: HeteroSVD
    • Data Input: Input DTable of user-item rating matrix data.
    • Model Output: Hetero SVD model.
    5. Hetero SVD++

    Build hetero SVD++ module through multiple parties.

    • Corresponding module name: HeteroSVDPP
    • Data Input: Input DTable of user-item rating matrix data.
    • Model Output: Hetero SVD++ model.
    6. Hetero GMF

    Build hetero GMF module through multiple parties.

    • Corresponding module name: HeteroGMF
    • Data Input: Input DTable of user-item rating matrix data(using positive data only).
    • Model Output: Hetero GMF model.

    More available algorithms are coming soon.

    转自https://github.com/FederatedAI/FATE

  • 相关阅读:
    C# 数组 随机 排序
    安全防护之加盐慢哈希加密
    NLog的介绍使用
    xmlns:xsi=”http://www.w3.org/2001/XMLSchema-instance(xsi:schemaLocation详解)
    如何计算时间复杂度(转)
    ppp协议介绍(转)
    Netlink 介绍(译)
    TIME_WAIT状态的一些总结
    带头结点单链表的翻转(递归)
    压缩前端文件(html, css, js)
  • 原文地址:https://www.cnblogs.com/lhwblogs/p/14729564.html
Copyright © 2011-2022 走看看