Layer that normalizes its inputs
layer_batch_normalization(
object,
axis = -1L,
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 = NULL,
gamma_regularizer = NULL,
beta_constraint = NULL,
gamma_constraint = NULL,
synchronized = FALSE,
...
)
Layer or model object
Integer, the axis that should be normalized (typically the features
axis). For instance, after a Conv2D
layer with
data_format="channels_first"
, set axis=1
in BatchNormalization
.
Momentum for the moving average.
Small float added to variance to avoid dividing by zero.
If TRUE
, add offset of beta
to normalized tensor. If FALSE
,
beta
is ignored.
If TRUE
, multiply by gamma
. If FALSE
, gamma
is not used. When
the next layer is linear (also e.g. nn.relu
), this can be disabled
since the scaling will be done by the next layer.
Initializer for the beta weight.
Initializer for the gamma weight.
Initializer for the moving mean.
Initializer for the moving variance.
Optional regularizer for the beta weight.
Optional regularizer for the gamma weight.
Optional constraint for the beta weight.
Optional constraint for the gamma weight.
If TRUE
, synchronizes the global batch statistics (mean and
variance) for the layer across all devices at each training step in a
distributed training strategy. If FALSE
, each replica uses its own
local batch statistics. Only relevant when used inside a
tf$distribute
strategy.
standard layer arguments.
Batch normalization applies a transformation that maintains the mean output close to 0 and the output standard deviation close to 1.
Importantly, batch normalization works differently during training and during inference.
During training (i.e. when using fit()
or when calling the layer/model
with the argument training=TRUE
), the layer normalizes its output using
the mean and standard deviation of the current batch of inputs. That is to
say, for each channel being normalized, the layer returns
gamma * (batch - mean(batch)) / sqrt(var(batch) + epsilon) + beta
, where:
epsilon
is small constant (configurable as part of the constructor
arguments)
gamma
is a learned scaling factor (initialized as 1), which
can be disabled by passing scale=FALSE
to the constructor.
beta
is a learned offset factor (initialized as 0), which
can be disabled by passing center=FALSE
to the constructor.
During inference (i.e. when using evaluate()
or predict()
or when
calling the layer/model with the argument training=FALSE
(which is the
default), the layer normalizes its output using a moving average of the
mean and standard deviation of the batches it has seen during training. That
is to say, it returns
gamma * (batch - self.moving_mean) / sqrt(self.moving_var+epsilon) + beta
.
self$moving_mean
and self$moving_var
are non-trainable variables that
are updated each time the layer in called in training mode, as such:
moving_mean = moving_mean * momentum + mean(batch) * (1 - momentum)
moving_var = moving_var * momentum + var(batch) * (1 - momentum)
As such, the layer will only normalize its inputs during inference after having been trained on data that has similar statistics as the inference data.
When synchronized=TRUE
is set and if this layer is used within a
tf$distribute
strategy, there will be an allreduce
call
to aggregate batch statistics across all replicas at every
training step. Setting synchronized
has no impact when the model is
trained without specifying any distribution strategy.
Example usage:
strategy <- tf$distribute$MirroredStrategy()with(strategy$scope(), {
model <- keras_model_sequential()
model %>%
layer_dense(16) %>%
layer_batch_normalization(synchronized=TRUE)
})