Investigating Data Distribution Variability Across Devices in Federated Learning: Comparative Analysis of Algorithm Performance
Duda, Leonhard 1; Alibabaei, Khadijeh
1; Berberi, Lisana
1; Kozlov, Valentin
1; Streit, Achim
1
1 Scientific Computing Center (SCC), Karlsruher Institut für Technologie (KIT)
1; Berberi, Lisana
1; Kozlov, Valentin
1; Streit, Achim
11 Scientific Computing Center (SCC), Karlsruher Institut für Technologie (KIT)
Abstract:
Federated Learning (FL) enables distributed training on multiple devices, enhancing privacy and conserving resources by sharing model updates instead of data. Using NVFlare, we distributed the training of a CNN for brain tumor detection, keeping sensitive data local. We evaluated different FL algorithms (FedAvg, FedOpt, FedProx, Scaffold) and found that complex algorithms like Scaffold perform better among heterogeneous data distributions.
| Zugehörige Institution(en) am KIT | Scientific Computing Center (SCC) |
| Publikationstyp | Poster |
| Publikationsdatum | 12.06.2024 |
| Sprache | Englisch |
| Identifikator | KITopen-ID: 1000171684 |
| HGF-Programm | 46.21.02 (POF IV, LK 01) Cross-Domain ATMLs and Research Groups |
| Veranstaltung | Helmholtz Artificial Intelligence Conference (Helmholtz AI 2024), Düsseldorf, Deutschland, 12.06.2024 – 14.06.2024 |
| Projektinformation | AI4EOSC (EU, EU 9. RP, 101058593) |
| Schlagwörter | Federated Learning |