Major advancements in artificial intelligence (AI) combined with the constantly increasing computing power pushed adoption of AI in our everyday life. However, the learning process requires to collect and centralize user’s data in datacenters outside of the user’s control. This practice raises some concerns from users who fear the misuse of their data. On the other side, some use cases cannot be explored as it requires data that is forbidden to share. For example, financial or medical data. To remedy to these problems, a federated approach was suggested. In this approach, users train a model locally on their device and only have to share the global model with a centralized server. Its role is to aggregate models from all users. This approach inspired a lot new variants.
This paper proposes a new decentralised and inclusive version. It recognizes the sovereignty of the different actors and gives total control on the training and aggregation of models. This document document details the architecture, the implementation and the results of this approach.
| Date | 16 Aug 2023 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Kaiwen Zhang (Supervisor) |
|---|
Duchesne, M. (Author),
Zhang (Supervisor),
16 Aug 2023Student thesis: Master's thesis › Master in Engineering: Engineering