FedGenTeacherLoss
Enums · AiDotNet.Enums
Selects the teacher term of the FedGen server generator objective (Zhu et al. 2021).
For Beginners
The server generator is trained so that the clients’ classifiers recognise the features it produces. These options choose how the clients’ opinions are combined into that training signal.
Fields
| Field | Summary |
|---|---|
FedGenTeacherLoss EnsembleCrossEntropy |
Cross-entropy of the label-count-weighted ensemble of client predictor logits, CE(Σ_k α_k^y h_k(z), y) (the paper’s generator objective). |
FedGenTeacherLoss WeightedClientCrossEntropy |
Label-count-weighted sum of the clients’ own cross-entropies, Σ_k α_k^y CE(h_k(z), y) (the teacher loss of the authors’ released code, serverpFedGen.train_generator). |
int value__ |