Glossary term
Glossary term
Safety and Alignment
A common approach to self-supervised learning in which:
Noise is artificially added to the dataset.
The model tries to remove the noise.
Denoising enables learning from unlabeled examples. The original dataset serves as the target or label and the noisy data as the input.
Some masked language models use denoising as follows:
Noise is artificially added to an unlabeled sentence by masking some of the tokens.
The model tries to predict the original tokens.
Created for this library
An OCR vendor trains a denoising preprocessor on photographs of receipts so its extraction model is robust to creased and shadowed images.
A medical imaging team applies denoising to low-dose CT scans before downstream segmentation to preserve thin structures.
A speech recognition vendor applies a denoising front-end to call-center audio before passing it to the recognizer.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License