GaussianNoise layer
GaussianNoise class
keras.layers.GaussianNoise(stddev, seed=None, **kwargs)
Apply additive zero-centered Gaussian noise.
This is useful to mitigate overfitting (you could see it as a form of random data augmentation). Gaussian Noise (GS) is a natural choice as corruption process for real valued inputs.
As it is a regularization layer, it is only active at training time.
Arguments
- stddev: Float, standard deviation of the noise distribution.
- seed: Integer, optional random seed to enable deterministic behavior.
Call arguments
- inputs: Input tensor (of any rank).
- training: Python boolean indicating whether the layer should behave in training mode (adding noise) or in inference mode (doing nothing).