FederatedPersonalizationStrategy
Enums · AiDotNet.Models.Options
Specifies the personalization strategy for federated learning.
For Beginners
Each strategy determines how clients adapt the global model to their local data.
Fields
| Field | Summary |
|---|---|
FederatedPersonalizationStrategy Clustered |
Clustered — cluster clients by gradient similarity, aggregate within clusters. |
FederatedPersonalizationStrategy Ditto |
Ditto — train a regularized personalized model alongside the global model. |
FederatedPersonalizationStrategy FedAGHN |
FedAGHN — adaptive gradient-based heterogeneous networks. |
FederatedPersonalizationStrategy FedBABU |
FedBABU — freeze head during FL, fine-tune body, then locally fine-tune head. |
FederatedPersonalizationStrategy FedCP |
FedCP — conditional computation policy routing inputs to model subsets. |
FederatedPersonalizationStrategy FedPAC |
FedPAC — personalization via aggregation and calibration with prototype alignment. |
FederatedPersonalizationStrategy FedPer |
FedPer — personalize the last (classification head) layers, share the body. |
FederatedPersonalizationStrategy FedRep |
FedRep — learn shared representations with personalized heads; alternating optimization. |
FederatedPersonalizationStrategy FedRoD |
FedRoD — dual classifiers: one aggregated generic + one local personalized. |
FederatedPersonalizationStrategy FedSelect |
FedSelect — learned sparse binary masks determining personalized vs shared params. |
FederatedPersonalizationStrategy KNNPer |
kNN-Per — kNN cache over global features for zero-cost personalization at inference. |
FederatedPersonalizationStrategy None |
No personalization — all parameters are aggregated globally. |
FederatedPersonalizationStrategy PFedGate |
pFedGate — gated layer-wise mixture of local and global parameters. |
FederatedPersonalizationStrategy PFedMe |
pFedMe — Moreau-envelope-based personalization with proximal local solver. |
int value__ |