Denoising Autoencoder (DAE)

Definition - What does Denoising Autoencoder (DAE) mean?

A denoising autoencoder is a specific type of autoencoder, which is generally classed as a type of deep neural network. The denoising autoencoder gets trained to use a hidden layer to reconstruct a particular model based on its inputs.

Techopedia explains Denoising Autoencoder (DAE)

In general, autoencoders work on the premise of reconstructing their inputs. Autoencoders are generally unsupervised machine learning programs deriving results from unstructured data.

To achieve this equilibrium of matching target outputs to inputs, denoising autoencoders accomplish this goal in a specific way – the program takes in a corrupted version of some model, and tries to reconstruct a clean model through the use of denoising techniques. Engineers may apply noise in a particular amount as a percentage of the model and try to force the hidden layer to work from the corrupted version to produce a clean version. Denoising autoencoders can also be stacked on each other to provide iterative learning toward this key goal.

Share this: