Question about vertical federated learning

[jiahuanluo]

Thanks for your job!
But I am confused about the concept of vertical federated learning in your paper.
Therefore, the server has access to both of the model parameters and their gradients.
In vertical federated learning, why should the server access the local model parameters?

[DeRafael]

Hi thank you for your question.
The VFL protocol we applied in our paper is from paper "Federated Machine Learning: Concept and Applications" which is a typical and widely used VFL training protocol
In their proposed protocol, the server is the one responsible for model parameter updating and broadcasting. This is the reason why the server has the access to both the model parameter and the gradients of loss w.r.t them.

You are welcomed to send me questions through the email jinxiao96@gmail.com which I will response you quicker.

Best,
Xiao Jin