The proposed Advance FL Ensemble Algorithm
Required: Annotated trained local models
Ensure: Decision via advanced FL ensemble method
Initialization: Assign initial weights to samples
for
1. Train the classifier using the weighted training local sets
2. Evaluate the error of the classifier with:
3. Determine classifier weightusing a complexity penalty factor λ:
4. Update sample weights:
5. Normalize Weight: end for
FL Ensemble Model Decision:
1. Combine local model classifiers using a non-linear aggregation rule: For each class compute: where is a dynamic scalar factor for the classifier’s vote.
2. Assign the test sampleto the class with