On the Inherent Regularization Effects of Noise Injection During Training.
Oussama DhifallahYue M. LuPublished in: ICML (2021)
Keyphrases
- early stopping
- training process
- random noise
- additive noise
- image noise
- noisy data
- neural network
- arbitrary shape
- signal to noise ratio
- missing data
- test set
- online learning
- image deconvolution
- training and testing data
- impulse noise removal
- noise free
- train a neural network
- noise removal
- gaussian noise
- training phase
- regularization parameter
- noise level
- noise reduction
- maximum likelihood
- high resolution
- training set
- multiscale
- image processing
- machine learning