FlowBasedDistillationStrategy<T>

Models & Types · AiDotNet.KnowledgeDistillation.Strategies

Flow-based distillation that matches the information flow between layers.

Inheritance: DistillationStrategyBase<T> → FlowBasedDistillationStrategy<T>

Constructors

Constructor Summary
FlowBasedDistillationStrategy(double flowWeight, double temperature, double alpha)

Methods

Method Summary
Matrix<T> ComputeGradient(Matrix<T> studentBatchOutput, Matrix<T> teacherBatchOutput, Matrix<T> trueLabelsBatch)
T ComputeFlowLoss(Tensor<T>[] studentFeatures, Tensor<T>[] teacherFeatures) Computes the FSP (“flow of solution procedure”) loss from structured feature MAPS, which is the form the paper defines.
T ComputeFlowLoss(Vector<T>[] studentFeatures, Vector<T>[] teacherFeatures) Computes the FSP (“flow of solution procedure”) loss from structured feature MAPS, which is the form the paper defines.
T ComputeLoss(Matrix<T> studentBatchOutput, Matrix<T> teacherBatchOutput, Matrix<T> trueLabelsBatch)