Federated Machine Learning

What is Federated Learning:

Federated learning (FL) is a machine learning system. We have many clients (e.g. mobile devices or whole organizations) collaboratively train a model under the infrastructure of a central server (e.g. service provider) while keeping the training data decentralized.

We also have some other concepts:

Enable machine learning engineers and data scientists to work productively with decentralized data with privacy by default (*Google*) or Data does not move, only algorithms travel (*owkin.com*)

The Research Directions about Federated Learning

References at: https://github.com/haithienld/ListFLPaper

The Lifecycle of a Model in Federated Learning.

The life cycle of a Model in Federated Learning is described as bellow image:

• 1. Problem Definition and Client selection.

• 2. Broadcast the tasks to the devices which have the same problem to solve.

• 3. Client computation. (Training local data set at the client).

• 4.5.6 send back the trained model to the server.

• 7. Average Model or Model Aggregation.

• 8. Model update.

• 9. Deploy a new model.

Motivation:

As we described above, there are many FL research directions. However, We focus on the privacy and security problems of FL due to these are the features that distinguish Federated Learning and Distributed Learning.

Besides that, our objective is to develop an end-to-end FL platform that can serve different purposes related to many fields such as whole organizations (hospitals, state agencies), or mobile system training.

Propose architecture

Our proposed architecture is described in the following figure:

Implementation Roadmap

If you any question, you can contact to: haithienld@snu.ac.kr