IVerticalFederatedTrainer<T>

Interfaces · AiDotNet.FederatedLearning.Vertical

Orchestrates the vertical federated learning training process.

Known implementations: VerticalFederatedTrainer<T>

For Beginners

This is the main coordinator for vertical FL training. It handles the entire pipeline:

How It Works

Unlike horizontal FL trainers (which average model parameters), the VFL trainer coordinates activations flowing through a split neural network. No party ever sees another party’s raw features.

Methods

Method Summary
Tensor<T> Predict(IReadOnlyList<int> entityIndices) Makes predictions on aligned entity data using the split model.
VflAlignmentSummary AlignEntities(PsiOptions psiOptions) Performs entity alignment across all registered parties using PSI.
VflEpochResult<T> TrainEpoch() Runs a single training epoch over the aligned data.
VflTrainingResult<T> Train() Runs the full training loop for the configured number of epochs.
int UnlearnEntities(IReadOnlyList<string> entityIds) Removes the influence of specified entities from the trained model (unlearning).
void RegisterParty(IVerticalParty<T> party) Registers a party to participate in VFL training.