Design, Implementation, and Analysis of a Federated Learning Architecture
CovenanteeKing's College London
Document typeMaster thesis
Rights accessOpen Access
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder
Artificial Intelligence (AI) has become the go-to technology for many applications. This method, even having huge advantages in many areas, also require from training step, which is very power-and-time consuming and has some privacy concerns. As a consequence, distributed training techniques are emerging, such as Federated Learning (FL), because it does not only enables to distribute the computational load but training sensible data locally without the necessity of sending it to a main server, for privacy reasons. In this thesis, we present a framework that achieves exactly this, tested with tactile data, which as far as we are aware, has never been tried before. The framework is lightweight and uses encrypted HTTP protocol to exchange information between the different parties. It is benchmarked in different scenarios, such as same-server and multiple-server configurations, and some results show how it can be superior/faster to the centralized scenario, with the plus of more robust security.
Study the effects of the asynchrony of heterogeneous devices in Federated Learning Systems. Understanding asynchrony as the global aggregations not done simulteanously in different nodes, while heterogeneous devices are those having different computation power in different nodes.
DegreeMÀSTER UNIVERSITARI EN TECNOLOGIES AVANÇADES DE TELECOMUNICACIÓ (Pla 2019)