Overview

Mon, 01 Jan 0001 00:00:00 +0000

Federated Learning (FL) distributes the computational load of machine learning while preserving privacy by keeping sensitive data on the client devices instead of a centralized server.

FL Basics

Explore fundamental concepts of Federated Learning

FL Basics

Mon, 01 Jan 0001 00:00:00 +0000

Before we discuss Federated Learning (FL) in detail, let’s recall how classic centralized machine learning (ML) works.

Classic Machine Learning

The classic centralized ML model training process can be represented as follows: Initially, each client has its data (D). The centralized server holds an untrained (gray) ML model (M).

Data is required to enable model training, which clients must send to the server.

After/during training, the client's data remains on the server and is exposed to potential exploitation and privacy breaches.

( The pink/purple model color symbolizes different data sources used during training. )

Federated Learning Overview

In FL, the server/aggregator coordinates the FL processes but does not train the model. Each client/learner does the training locally, which requires setting up an environment to handle training. All FL components must know and possess the local ML model, which is initially untrained.

The aggregator starts the first FL training cycle. Each selected learner starts training its model copy exclusively using local data.

Naming Conventions

Federated Machine Learning on Oakestra

Overview

FL Basics

FL Basics

Classic Machine Learning

Federated Learning Overview