Federated Learning is proposed by Google in 2016. In this project, we try to deploy a news recommendation algorithm on a federated framework - Pysyft.

Github: https://github.com/ZHANG-GuiGui/news_reco_syft/

CSDN: https://blog.csdn.net/weixin_45520982/article/details/116781765/

Data security and personal privacy are two main concerns with regard to giant Internet companies.

Federated Learning is about how to update the deep learning model stored on a central server without seeing training data conserved on multiple client devices.

The simplest way is to transfer encrypted gradient calculated locally on clients' devices.

Our project is more like an engineer attempt without considering fondly the theories behind Federated Learning.

We used the MIcrosoft News Dataset (MIND). Due to project time limitation, we conducted the experiments on MIND-small dataset.

We used the NRMS model as the default news recommendation model.

Pysyft 0.2.x is chosen as our deploy framework.

• batch size = 128
• learning rate = 5e-4 for Adam Optimizer 0.01 for SGD Optimizer

Team Member:

HUANG Zeyu

ZHANG Xiaofeng

LI Yuhang

WANG Kunyang

TAN Ying

Supervisor:

JIANG Li @ Orange Beijing Lab