Enum FederatedStalenessWeighting
Specifies how to down-weight stale updates in asynchronous federated learning.
public enum FederatedStalenessWeighting
Fields
Constant = 0No additional staleness weighting (constant weight).
Exponential = 2Weight = exp(-rate * staleness).
Inverse = 1Weight = 1 / (1 + staleness).
Polynomial = 3Weight = 1 / (1 + staleness)^rate.
Remarks
For Beginners: If an update was computed on an old global model, it can be less helpful. Staleness weighting reduces the influence of older updates.