Fig 1.
Communication between the clients and the server under FedAvg.
Fig 2.
FedAvg complemented by the data-sharing strategy: Distribute shared data to the clients at initialization.
Fig 3.
Communication between the clients and the server under LoAdaBoost FedAvg.
Table 1.
Summary of the evaluation dataset.
Table 2.
Example rows and columns of DRUGS.
Fig 4.
Performance gap between IID and non-IID data.
Fig 5.
Comparison of FedAvg and LoAdaboost on IID data.
LoAdaBoost converged slightly slower than FedAvg, but to a higher test AUC.
Table 3.
IID scenario: 10-fold cross validation results with varying C and E.
Fig 6.
Comparison of FedAvg and LoAdaboost on non-IID data with data-sharing strategy.
Table 4.
Non-IID scenario: 10-fold cross validation results with varying α and β.
Table 5.
Non-IID scenario: 10-fold cross validation results with varying C.
Table 6.
Summary of the eICU dataset.
Fig 7.
Comparison of FedAvg and LoAdaboostFedAvg on eICU data.
Table 7.
Evaluation on eICU data: 10-fold cross validation results.