Result Page - Federated Learning Testbed Project
The Vinnova-financed Federated Learning Testbed project has been running from August 2020 till June 2021, where the purpose of the project has been to create a virtual testbed to build up a collective know-how of how to work with this technology in a production setting and investigate the legal and privacy requirements.
Below is a list of relevant results and learnings from the project. Some resources and links will be updated as soon as the documents or content is ready. Revisit this page continuously or send us an email
List of learnings and results
- How to get started with Federated Learning - Created the following blogpost with our general learnings from working with Federated Learning for the past year. (Currently being finalized. Coming asap. Updated 2021-05-27))
- Legal Reasoning Document - An investigation and discussion of how current legislation and GDPR relate to models being trained in a federated setting. Relevant for both legal and technical people at organizations who are about to start testing and using federated Learning. (For everyone. Soon to come (updated 2021-05-27)) (Currently being finalized. Coming June 2021)
- FedBird Demo - Try Federated Learning yourself in a cross-silo setting, using Scaleout’s open-source framework called FEDn. The demo trains an object detection model on 2x nodes, where each node contains a different portion of the data set isolated from one another. You can access the interactive demo here https://172.25.16.5:8090/ if you have access to the edge lab. Demo-walkthrough can be seen on YouTube here. (The demo is only for AI Sweden members due to HW and dataset agreements. If you are an AI Sweden partner and interested in testing for yourself, send us an email.)
- An overview and benchmark of useful FL-frameworks - A whitepaper from one of AI Sweden’s partners, Smartilizer. The paper presents an overview of some of the most common available federated learning frameworks, where some of them have been rigorously tested in performance. This paper from 2021-05-26 is available for AI Sweden partners as part of the membership in AI Sweden. For non-partners, the paper can be purchased at a cost of 35 kSEK. If interested, send us an email.
- Edge Lab - An Edge Learning infrastructure that opened January 2021 where Federated Learning and decentralized AI frameworks and algorithms, can be tested in a safe and as industry-like environment as possible. (Available for AI Sweden partners only.)
- Official Vinnova Project report - A project is not a project without a report. The same goes for this project. (Coming June 2021)
About
Partners and contributors to the testbed project have been: Zenseact, Scaleout Systems, RISE, Peltarion, MAQS Law firm, Qamcom, HPE, Fraunhofer Chalmers Research Centre for Industrial Mathematics (FCC), Halmstad University Region Halland, and Smartilizer.
