LayerAttributionExplainer<T>
Models & Types · AiDotNet.Interpretability.Explainers
Layer-level attribution explainer for computing attributions at intermediate layers. Supports LayerIntegratedGradients, LayerDeepLIFT, LayerGradientXActivation, and LayerConductance.
Implements: IGPUAcceleratedExplainer<T>
For Beginners
While input attribution tells you which input features matter, layer attribution tells you which NEURONS IN A HIDDEN LAYER matter for the output.
Why layer attribution?
- Higher-level features: Later layers encode abstract concepts, not raw pixels
- Model understanding: See what the model “thinks” at each stage
- Debugging: Find where information is lost or distorted
- Transfer learning: Understand which learned representations are being used
Comparison of methods:
- LayerGradient: Simple ∂output/∂layer, fast but noisy
- LayerIntegratedGradients: Path-integrated gradients, theoretically grounded
- LayerDeepLIFT: Difference from reference, handles saturation well
- LayerConductance: Layer activation × gradient, captures both magnitude and sensitivity
Example: For an image classifier, layer attribution on the last conv layer might show that specific feature maps (detecting eyes, fur, etc.) are important for the “cat” prediction.
Constructors
| Constructor | Summary |
|---|---|
LayerAttributionExplainer(Func<Vector<T>, Vector<T>> predictFunction, Func<Vector<T>, Vector<T>> layerActivationFunction, Func<Vector<T>, int, Vector<T>> layerGradientFunction, int layerSize, int inputSize, LayerAttributionMethod method, int integrationSteps) |
Initializes a layer attribution explainer. |
Properties
| Property | Summary |
|---|---|
bool IsGPUAccelerated { get; } |
|
bool SupportsGlobalExplanations { get; } |
Gets whether this explainer supports global explanations. |
bool SupportsLocalExplanations { get; } |
Gets whether this explainer supports local explanations. |
string MethodName { get; } |
Gets the method name. |
Methods
| Method | Summary |
|---|---|
LayerAttributionResult<T> ComputeLayerAttribution(Vector<T> instance, int? targetClass) |
Computes layer attribution for the specified layer. |
void SetGPUHelper(GPUExplainerHelper<T> helper) |