tf.keras.layers.BatchNormalization

View source on GitHub

Normalize and scale inputs or activations. (Ioffe and Szegedy, 2014).

tf.keras.layers.BatchNormalization(
    axis=-1, momentum=0.99, epsilon=0.001, center=True, scale=True,
    beta_initializer='zeros', gamma_initializer='ones',
    moving_mean_initializer='zeros', moving_variance_initializer='ones',
    beta_regularizer=None, gamma_regularizer=None, beta_constraint=None,
    gamma_constraint=None, renorm=False, renorm_clipping=None, renorm_momentum=0.99,
    fused=None, trainable=True, virtual_batch_size=None, adjustment=None, name=None,
    **kwargs
)

Normalize the activations of the previous layer at each batch, i.e. applies a transformation that maintains the mean activation close to 0 and the activation standard deviation close to 1.

Batch normalization differs from other layers in several key aspects:

1) Adding BatchNormalization with training=True to a model causes the result of one example to depend on the contents of all other examples in a minibatch. Be careful when padding batches or masking examples, as these can change the minibatch statistics and affect other examples.

2) Updates to the weights (moving statistics) are based on the forward pass of a model rather than the result of gradient computations.

3) When performing inference using a model containing batch normalization, it is generally (though not always) desirable to use accumulated statistics rather than mini-batch statistics. This is acomplished by passing training=False when calling the model, or using model.predict.

Arguments:

Call arguments:

Input shape:

Arbitrary. Use the keyword argument input_shape (tuple of integers, does not include the samples axis) when using this layer as the first layer in a model.

Output shape:

Same shape as input.

About setting layer.trainable = False on a `BatchNormalization layer:

The meaning of setting layer.trainable = False is to freeze the layer, i.e. its internal state will not change during training: its trainable weights will not be updated during fit() or train_on_batch(), and its state updates will not be run.

Usually, this does not necessarily mean that the layer is run in inference mode (which is normally controlled by the training argument that can be passed when calling a layer). "Frozen state" and "inference mode" are two separate concepts.

However, in the case of the BatchNormalization layer, setting trainable = False on the layer means that the layer will be subsequently run in inference mode (meaning that it will use the moving mean and the moving variance to normalize the current batch, rather than using the mean and variance of the current batch).

This behavior has been introduced in TensorFlow 2.0, in order to enable layer.trainable = False to produce the most commonly expected behavior in the convnet fine-tuning use case.

Note that:

Normalization equations: Consider the intermediate activations (x) of a mini-batch of size (m):

We can compute the mean and variance of the batch

({\mu_B} = \frac{1}{m} \sum_{i=1}{m} {x_i})

({\sigma_B2} = \frac{1}{m} \sum_{i=1}{m} ({x_i} - {\mu_B})2)

and then compute a normalized (x), including a small factor ({\epsilon}) for numerical stability.

(\hat{x_i} = \frac{x_i - \mu_B}{\sqrt{\sigma_B2 + \epsilon}})

And finally (\hat{x}) is linearly transformed by ({\gamma}) and ({\beta}), which are learned parameters:

({y_i} = {\gamma * \hat{x_i} + \beta})

References: