BeyondFederated Truly decentralised edge AI

Master Thesis (2024)
Author(s)

Q. van Eijs (TU Delft - Electrical Engineering, Mathematics and Computer Science)

Contributor(s)

Johan Pouwelse – Mentor (TU Delft - Data-Intensive Systems)

Faculty
Electrical Engineering, Mathematics and Computer Science
More Info
expand_more
Publication Year
2024
Language
English
Graduation Date
13-06-2024
Awarding Institution
Delft University of Technology
Programme
['Computer Science']
Faculty
Electrical Engineering, Mathematics and Computer Science
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

Decentralized learning (DL) leverages edge devices for collaborative model training while enhancing data privacy, as training data never leaves the device. In this paper, we present "BeyondFederated: Truly Decentralised Edge AI," a novel approach to overcoming the limitations of traditional Federated Learning (FL) by eliminating reliance on centralized servers. Traditional FL, despite its decentralization benefits, still poses significant privacy and security risks due to its central control points. Our system leverages edge AI with TensorFlow Lite and a peer-to-peer gossip network to ensure fully decentralized learning and data processing. We developed a proof-of-concept to demonstrate a decentralized alternative to Spotify, incorporating Web3 YouTube playback, and enabling efficient and accurate Scalable Nearest Neighbor (ScaNN) searches using vector embeddings incorporated into an actual TensorFlow model running on edge devices. Through various experiments, we evaluated the system's performance in real-time search capabilities, embedding model comparisons, and new data insertion handling. The results confirm that BeyondFederated maintains high efficiency, scalability, and privacy. This study underscores the potential for truly decentralized machine learning and sets the stage for more robust and user-centric decentralized AI applications.

Files

MSCThesisQuintenVanEijs.pdf
(pdf | 1.88 Mb)
License info not available