Federated Learning over a Multi-Agent Framework (FL-MAS)
SONAR|HES-SO
- Meichtry, Joel
- Schumacher, Michael Ignaz (Degree supervisor)
- Sierre : Haute Ecole de Gestion Valais, 2024
96 pages
Bachelor of Science HES-SO (BSc) in Business Information Technology: Haute Ecole de Gestion Valais, 2024
English
This bachelor’s thesis was suggested by Davide Calvaresi. The first goal of this bachelor thesis was to implement federated learning with a multi-agent system. The second goal is that the implementation allows for the dynamic entry of an agent. This agent should also be capable of directly contributing after he joins. The third goal is that the agents can identify fashion objects that weren’t present in their previous dataset.
The first goal was implemented with the help of the machine learning framework PyTorch and the multi-agent system platform Smart Python Agent Development Environment. The results show that dynamic entry by an agent is possible. The newly added agent could also directly contribute to the federated learning process after he joined. The result also showed that it wasn’t possible to implement the third goal. The agents could only identify unknown fashion objects, that weren’t in the original dataset for a short period of time.
The first goal was implemented with the help of the machine learning framework PyTorch and the multi-agent system platform Smart Python Agent Development Environment. The results show that dynamic entry by an agent is possible. The newly added agent could also directly contribute to the federated learning process after he joined. The result also showed that it wasn’t possible to implement the third goal. The agents could only identify unknown fashion objects, that weren’t in the original dataset for a short period of time.
- Language
-
- English
- Classification
- Computer science and technology
- Notes
-
- Haute Ecole de Gestion Valais
- Informatique de gestion - Wirtschaftsinformatik
- hesso:hegvs
- Persistent URL
- https://sonar.ch/global/documents/329926
Statistics
Document views: 56
File downloads:
- Meichtry_Joel_2024.pdf: 97