Denoising Autoencoder

The basic principle of the noise reduction autoencoder is as follows:

1. Add noise to the input data.
2. Compress noisy input data into a low-dimensional representation with an encoder.
3. A decoder is used to restore the low-dimensional representation to the original data.
4. The network is trained so that the output of the decoder is as close as possible to the original data, while the encoder learns to learn features from the noise.

The purpose of a denoising autoencoder is to learn how to recover clean output data from corrupted input data.

1. Image Denoising. Denoising autoencoders can be used to remove Gaussian or uniform noise from images.
2. Data preprocessing. In some machine learning tasks, due to the presence of noise or missing data, it is necessary to preprocess the input data. At this time, a denoising autoencoder can be used to extract a clean feature representation.
3. Visual feature representation. Denoising autoencoders can learn low-dimensional representations of noise-free input data, and these low-dimensional representations can be used for tasks such as image generation and data classification.

Original reference https://www.python100.com/html/98VA526PT8WW.html