IAdversarialInputPreprocessor<T>

Interfaces · AiDotNet.Interfaces

An input transformation applied before inference to remove adversarial perturbations.

For Beginners

This is a “cleaning step” your inputs go through before the model sees them.

How It Works

Implement this to plug a custom denoiser (for example a learned autoencoder, or a high-level representation guided denoiser) into AdversarialTraining through CustomPreprocessor. A configured preprocessor replaces the built-in AdversarialPreprocessingMethod.

Methods

Method Summary
Vector<T> Preprocess(Vector<T> input) Transforms one flattened input.