FEDn

FEDn – A scalable federated machine learning framework for cross-device
and cross-silo environments

Keywords
Scalability; Efficiency; Cybersecurity; Attack landscape

Intended audience
Intermediate and advanced

Abstract
Federated machine (FL) learning has opened new avenues for privacy-preserving data analysis.
The classical implementation of federated learning has several limitations, including restricted
scalability, efficiency concerns, and vulnerabilities to cyberattacks. Recent research has shown
that due to the distributed nature of the whole training process in FL, the previously reported
impact of attacks on centralized model training cannot be directly mapped to federated model
training. The attack landscape needs to be adapted for FL. This Workshop will cover the
architectural details of FEDn, a framework designed to address the scalability, efficiency and
security limitations of federated machine learning. We will discuss how FEDn effectively supports
both cross-device and cross-silo use cases. We will also cover our ongoing efforts to
understand the impact of cyberattacks on federated learning and potential mitigation
strategies. The workshop will also cover a hands-on session where we jointly train a machine
learning model using FEDn studio.

Date and duration
18th September 2024, 2 hours

Presenters
Fredrik Wrede: Head of Engineering Scaleout, PhD Scientific Computing, Uppsala, Sweden.
Leads the development of Scaleout software, including the FEDn framework. Strong background
in ML, distributed systems and cloud from both academic research as well as industry.

Viktor Valadi: Machine Learning Engineer at Scaleout, Masters at Chalmers university in
Computer Science, Gothenburg, Sweden. Background in Federated Learning Cyber Security
research relating to both privacy and robustness.

Prerequisite knowledge or skills required for attendees
Introductory level understanding of neural networks
Software
● CLI: Basic understanding of the command-line environments
● Python 3.10 or above: Intermediate-level Python programming skills
Hardware
● A laptop or a virtual machine (OS: Linux, Windows or MacOS)
● Memory requirement, around 2GB
● Storage requirement, 5GB.

Agenda
● Introduction (30 minutes)
○ Introduction to FEDn
○ Challenges related to federated machine learning
○ FEDn architecture
○ Results based on cross-silo and cross-device use cases
● FL and Cybersecurity (20 minutes)
● Hands-on Session (60 minutes)
● Summary and closing remarks (10 minutes)

Support material

● GitHub: https://github.com/scaleoutsystems/fedn
● YouTube channel: https://www.youtube.com/channel/UCZVv30LFXMQUOswNDKuQpNA
● FEDn: https://www.scaleoutsystems.com/framework
● Docs: https://fedn.readthedocs.io/en/stable/index.html
Relevant articles
1. M. Ekmefjord, A. Ait-Mlouk, S. Alawadi, M. Åkesson, P. Singh, O. Spjuth, S. Toor, A.
Hellander. Scalable federated learning with FEDn.
https://doi.org/10.1109/CCGrid54584.2022.00065
2. S. Alawadi, A. Ait-Mlouk, S. Toor, A. Hellander. Toward efficient resource utilization at edge
nodes in federated learning. https://doi.org/10.1007/s13748-024-00322-3
3. L. Ju; T. Zhang; S. Toor; A. Hellander. Accelerating Fair Federated Learning: Adaptive
Federated Adam. https://doi.org/10.1109/TMLCN.2024.3423648
4. Dolor Sit Amet
5. Valadi, V., Qiu, X., De Gusmão, P. P. B., Lane, N. D., & Alibeigi, M. (2023). {FedVal}: Different
good or different bad in federated learning. In 32nd USENIX Security Symposium (USENIX
Security 23) (pp. 6365-6380).
6. Garg, S., Jönsson, H., Kalander, G., Nilsson, A., Pirange, B., Valadi, V., & Östman, J. (2024).
Poisoning Attacks on Federated Learning for Autonomous Driving. arXiv preprint
arXiv:2405.01073.