chainer.functions.decorrelated_batch_normalization¶

chainer.functions.decorrelated_batch_normalization(x, *, groups=16, eps=2e-05, running_mean=None, running_projection=None, decay=0.9)[source]¶

Decorrelated batch normalization function.

It takes the input variable x and normalizes it using batch statistics to make the output zero-mean and decorrelated.

Parameters

x (Variable) – Input variable.
groups (int) – Number of groups to use for group whitening.
eps (float) – Epsilon value for numerical stability.
running_mean (N-dimensional array) – Expected value of the mean. This is a running average of the mean over several mini-batches using the decay parameter. If None, the expected mean is initialized to zero.
running_projection (N-dimensional array) – Expected value of the project matrix. This is a running average of the projection over several mini-batches using the decay parameter. If None, the expected projected is initialized to the identity matrix.
decay (float) – Decay rate of moving average. It is used during training.

Returns

The output variable which has the same shape as \(x\).

Return type

Variable

See: Decorrelated Batch Normalization