Hands-On Artificial Intelligence for IoT

Denoising autoencoders

A denoising autoencoder learns from a corrupted (noisy) input; we feed the encoder network the noisy input and the reconstructed image from the decoder is compared with the original denoised input. The idea is that this will help the network learn how to denoise an input. The network does not just make a pixel-wise comparison, instead, in order to denoise the image, the network is forced to learn the information of neighboring pixels as well.

Once the autoencoder has learned the encoded features y, we can remove the decoder part of the network and use only the encoder part to achieve dimensionality reduction. The dimensionally reduced input can be fed to some other classification or regression model.