FusedBatchNormGradV3

public final class FusedBatchNormGradV3

Gradient for batch normalization.

Note that the size of 4D Tensors are defined by either "NHWC" or "NCHW". The size of 1D Tensors matches the dimension C of the 4D Tensors.

Nested Classes

class FusedBatchNormGradV3.Options Optional attributes for FusedBatchNormGradV3

Public Methods

static <T extends Number, U extends Number> FusedBatchNormGradV3 <T, U>
create ( Scope scope, Operand <T> yBackprop, Operand <T> x, Operand <Float> scale, Operand <U> reserveSpace1, Operand <U> reserveSpace2, Operand <U> reserveSpace3, Options... options)
Factory method to create a class wrapping a new FusedBatchNormGradV3 operation.
static FusedBatchNormGradV3.Options
dataFormat (String dataFormat)
static FusedBatchNormGradV3.Options
epsilon (Float epsilon)
static FusedBatchNormGradV3.Options
isTraining (Boolean isTraining)
Output <U>
offsetBackprop ()
A 1D Tensor for the gradient with respect to offset.
Output <U>
reserveSpace4 ()
Unused placeholder to match the mean input in FusedBatchNorm.
Output <U>
reserveSpace5 ()
Unused placeholder to match the variance input in FusedBatchNorm.
Output <U>
scaleBackprop ()
A 1D Tensor for the gradient with respect to scale.
Output <T>
xBackprop ()
A 4D Tensor for the gradient with respect to x.

Inherited Methods

Public Methods

public static FusedBatchNormGradV3 <T, U> create ( Scope scope, Operand <T> yBackprop, Operand <T> x, Operand <Float> scale, Operand <U> reserveSpace1, Operand <U> reserveSpace2, Operand <U> reserveSpace3, Options... options)

Factory method to create a class wrapping a new FusedBatchNormGradV3 operation.

Parameters
scope current scope
yBackprop A 4D Tensor for the gradient with respect to y.
x A 4D Tensor for input data.
scale A 1D Tensor for scaling factor, to scale the normalized x.
reserveSpace1 When is_training is True, a 1D Tensor for the computed batch mean to be reused in gradient computation. When is_training is False, a 1D Tensor for the population mean to be reused in both 1st and 2nd order gradient computation.
reserveSpace2 When is_training is True, a 1D Tensor for the computed batch variance (inverted variance in the cuDNN case) to be reused in gradient computation. When is_training is False, a 1D Tensor for the population variance to be reused in both 1st and 2nd order gradient computation.
reserveSpace3 When is_training is True, a 1D Tensor for some intermediate results to be reused in gradient computation. When is_training is False, a dummy empty Tensor will be created.
options carries optional attributes values
Returns
  • a new instance of FusedBatchNormGradV3

public static FusedBatchNormGradV3.Options dataFormat (String dataFormat)

Parameters
dataFormat The data format for y_backprop, x, x_backprop. Either "NHWC" (default) or "NCHW".

public static FusedBatchNormGradV3.Options epsilon (Float epsilon)

Parameters
epsilon A small float number added to the variance of x.

public static FusedBatchNormGradV3.Options isTraining (Boolean isTraining)

Parameters
isTraining A bool value to indicate the operation is for training (default) or inference.

public Output <U> offsetBackprop ()

A 1D Tensor for the gradient with respect to offset.

public Output <U> reserveSpace4 ()

Unused placeholder to match the mean input in FusedBatchNorm.

public Output <U> reserveSpace5 ()

Unused placeholder to match the variance input in FusedBatchNorm.

public Output <U> scaleBackprop ()

A 1D Tensor for the gradient with respect to scale.

public Output <T> xBackprop ()

A 4D Tensor for the gradient with respect to x.