Interface IInputGradientComputable<T>

Namespace

AiDotNet.Interfaces

Assembly

AiDotNet.dll

Interface for models that support computing gradients with respect to input data.

public interface IInputGradientComputable<T>

Type Parameters

T

The numeric data type (e.g., float, double).

Remarks

This interface enables models to compute how their output changes with respect to input modifications. Unlike IGradientComputable<T, TInput, TOutput> which computes gradients for model parameters, this interface computes gradients for the input itself.

Use Cases:

• Adversarial Examples: Generate minimal perturbations that cause misclassification
• Saliency Maps: Visualize which input features most affect the output
• Input Attribution: Understand model predictions through input sensitivity
• Gradient Penalties: WGAN-GP and other regularization techniques

For Beginners: When training a model, we compute gradients to adjust the model's internal parameters (weights). This interface instead computes how sensitive the output is to changes in the input data.

For example, if you have an image classifier:

• Parameter gradients tell us how to adjust weights to improve accuracy
• Input gradients tell us which pixels, if changed, would most affect the prediction

This is essential for adversarial robustness testing - we can find the smallest image change that fools the classifier.

Methods

ComputeInputGradient(Tensor<T>, Tensor<T>)

Computes the gradient of the model output with respect to the input using tensor format.

Tensor<T> ComputeInputGradient(Tensor<T> input, Tensor<T> outputGradient)

Parameters

input Tensor<T>

The input tensor for which to compute gradients.

outputGradient Tensor<T>

The gradient tensor with respect to the output.

Returns

Tensor<T>

The gradient tensor with respect to the input.

ComputeInputGradient(Vector<T>, Vector<T>)

Computes the gradient of the model output with respect to the input.

Vector<T> ComputeInputGradient(Vector<T> input, Vector<T> outputGradient)

Parameters

input Vector<T>

The input for which to compute gradients.

outputGradient Vector<T>

The gradient with respect to the output (typically from a loss function).

Returns

Vector<T>

The gradient with respect to the input.

Remarks

This method performs backpropagation through the model to compute input gradients. The outputGradient represents how much we "care" about each output dimension, typically derived from a loss function.

For Adversarial Attacks: Set outputGradient to emphasize the target class (for targeted attacks) or the true class (for untargeted attacks), then use the returned input gradient to perturb the input in a direction that maximizes misclassification.