Interpolation can hurt robust generalization even when there is no noise.
Konstantin DonhauserAlexandru TifreaMichael AerniReinhard HeckelFanny YangPublished in: NeurIPS (2021)
Keyphrases
- image noise
- noisy environments
- greater robustness
- noise level
- imaging artifacts
- signal to noise ratio
- noise free
- salt pepper
- weak edges
- low pass filtering
- computationally efficient
- real time
- registration errors
- image interpolation
- geometric distortions
- random noise
- estimation error
- speech recognition
- neural network
- data sets