Computer scienceData scienceMachine learningData preprocessingFeature engineering

Autoencoders

Theory

Subtle difference

Report a typo

Both denoising and contractive AEs are aimed to enhance the model’s robustness, but there are a few fundamental differences between them.

Connect the statement to the autoencoder type.

Choose one option for each row
Denoising AEContractive AE
The emphasis is on letting the encoder be resistant to some perturbations of the input
Designed to be sensitive to the local input space, but insensitive to the variations in presentation
Introduce a penalty to the loss to learn a function that is insensitive to small variations in the input space

Create a free account to access the full topic