Federated Transfer Learning

Warning

This post is more than a year old. Information may be outdated.

Federated Transfer Learning (FTL) is a technique that can be applied when two datasets differ in their samples and feature space.

Consider
- a bank in China
- an e-commerce company in the United States
Due to geographical restrictions, the user groups of these two institutions have only a small intersection.
Because of their different businesses, there is only a tiny overlap in the feature space between the two parties.

To solve this problem, transfer learning techniques can create a common representation between the two feature spaces. This is done using limited standard sample sets to learn the joint representation, which can then be used to make predictions for samples with only one-sided features. FTL is an essential extension of existing federated learning systems because it deals with problems beyond existing algorithms' scope.