IDomainAdapter<T>
Interfaces · AiDotNet.TransferLearning.DomainAdaptation
Defines the interface for adapting models to reduce distribution shift between source and target domains.
Known implementations: CORALDomainAdapter<T>, MMDDomainAdapter<T>
For Beginners
Domain adaptation is like helping someone adjust to a new environment. Even when features are mapped correctly, the source and target domains might have different statistical properties (like different averages or variability). A domain adapter helps reduce these differences so a model trained on source data works better on target data.
How It Works
Think of it like adjusting your eyes when moving from a bright room to a dim room - the objects are the same, but your perception needs to adapt to the new lighting conditions.
Properties
| Property | Summary |
|---|---|
bool RequiresTraining { get; } |
Determines if the adapter requires training before use. |
string AdaptationMethod { get; } |
Gets the name of the adaptation method. |
Methods
| Method | Summary |
|---|---|
Matrix<T> AdaptSource(Matrix<T> sourceData, Matrix<T> targetData) |
Adapts source domain data to better match the target domain distribution. |
Matrix<T> AdaptTarget(Matrix<T> targetData, Matrix<T> sourceData) |
Adapts target domain data to better match the source domain distribution. |
T ComputeDomainDiscrepancy(Matrix<T> sourceData, Matrix<T> targetData) |
Computes the domain discrepancy (how different the domains are). |
void Train(Matrix<T> sourceData, Matrix<T> targetData) |
Trains the domain adapter if required. |