FedEdu: Facilitating FL in an Educational Context and Evaluating Different FL Scenarios

Supervisor: Turgud Valiyev

Author: Julian Weihs, Sebastian Wildervanck

Abstract

This Thesis aims to implement and test a new federated learning (FL) framework, called FedEdu, which is specifically designed for educational purposes. The proposed framework will allow students to easily implement FL projects, without having to deal with setting up communication between hardware and designing a pipeline from scratch. The architecture of FedEdu will be based on Kubernetes and KubeFlow. Sebastian will work on the implementation of the FedEdu framework, while Julian will work on evaluations. These will consist of testing the capabilities of the implemented framework as well as studying the effects of different methods and common pitfalls of FL using stock market data from Yahoo Finance as datasets. The plan is for the contents of this work to be usable as a resource for future courses on federated machine learning.