Abstract:
Federated Learning (FL) has arose as a groundbreaking approach in distributed
machine learning, enabling collaborative global model training while upholding
data privacy and security. This research highlights the importance of individual
node performance in the architecture of Federated Learning exploring its effects on
accuracy and effectiveness of the learning process.
To achieve optimal results in FL, the significance of node selection is
emphasized. The challenges imposed by different data distributions on nodes,
network connections, and computational capacities are comprehensively analyzed
along with the proposed methodology to address them. Moreover, the abstract digs
into trade-offs between nodes based on their high performance comparing their
local and global results.
By experimental results and evaluations, the results sheds light on how node
performance influences model accuracy using techniques like federated average
and federated weighted average along with federated automatic weight
optimization. The goal is to provide an understanding of node performance in the
underlying architecture of federated learning.
