January 2018
Intermediate to advanced
474 pages
10h 23m
English
These layers can be added to the model to prevent overfitting by adding noise; they are also known as regularization layers. These layers operate the same way as the Dropout() and ActivityRegularizer() layers in the core layers section.
| Layer name | Description |
|
GaussianNoise |
This layer applies additive zero-centered Gaussian noise to the inputs. |
| GaussianDropout | This layer applies multiplicative one-centered Gaussian noise to the inputs. |
| AlphaDropout | This layer drops a certain percentage of inputs, such that the mean and variance of the outputs after the dropout match closely with the mean and variance of the inputs. |
Read now
Unlock full access