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.


  • 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).