FederatedDriftMethod
Enums ยท AiDotNet.Models.Options
Specifies the drift detection method used in federated learning.
For Beginners
These methods detect when client data distributions change over time. Each has different strengths โ some catch sudden changes quickly (PageHinkley, DDM), while others are better at gradual drift (ADWIN). Model-based methods (GradientDivergence, WeightDivergence) detect drift through changes in training behavior rather than raw statistics.
Fields
| Field | Summary |
|---|---|
FederatedDriftMethod ADWIN |
ADWIN (Adaptive Windowing): maintains a variable-length window and detects drift when two sub-windows have significantly different means. |
FederatedDriftMethod DDM |
DDM (Drift Detection Method): monitors error rate and standard deviation. |
FederatedDriftMethod GradientDivergence |
Gradient divergence: detects drift by comparing gradient directions between rounds. |
FederatedDriftMethod PageHinkley |
Page-Hinkley test: sequential analysis for detecting change in the mean of a process. |
FederatedDriftMethod WeightDivergence |
Weight divergence: detects drift by measuring how much model weights change relative to historical patterns. |
int value__ |