MPSMatrixBatchNormalizationGradient
#import <MPSMatrixBatchNormalization.h>
Inherits MPSMatrixBinaryKernel.
(void) - setNeuronType:parameterA:parameterB:parameterC:
(MPSCNNNeuronType) - neuronType
(float) - neuronParameterA
(float) - neuronParameterB
(float) - neuronParameterC
(nonnull instancetype) - initWithDevice:
(void) -
encodeToCommandBuffer:gradientMatrix:inputMatrix:meanVector:varianceVector:gammaVector:betaVector:resultGradientForDataMatrix:resultGradientForGammaVector:resultGradientForBetaVector:
(nullable instancetype) - initWithCoder:device:
(nonnull instancetype) - copyWithZone:device:
NSUInteger sourceNumberOfFeatureVectors
NSUInteger sourceInputFeatureChannels
float epsilon
This depends on Metal.framework.
A kernel to compute the gradient of the batch normalization
operation.
A MPSMatrixBatchNormalizationGradient object
computes the results of backpropagating the gradients of a loss function
with respect to the outputs of an MPSMatrixBatchNormalization object.
The corresponding properties and data used by the
MPSMatrixBatchNormalizationGradient object should correspond to those
used by the forward MPSMatrixBatchNormalization object for which the
gradient is being computed.
- (nonnull instancetype) copyWithZone: (nullable NSZone *)
zone(nullable id< MTLDevice >) device
Make a copy of this kernel for a new device -
See also:
MPSKernel
Parameters:
zone The NSZone in which to allocate the object
device The device for the new MPSKernel. If nil, then use
self.device.
Returns:
A pointer to a copy of this MPSKernel. This
will fail, returning nil if the device is not supported. Devices must be
MTLFeatureSet_iOS_GPUFamily2_v1 or later.
Reimplemented from MPSKernel.
- (void) encodeToCommandBuffer: (nonnull id< MTLCommandBuffer
>) commandBuffer(MPSMatrix *__nonnull)
gradientMatrix(MPSMatrix *__nonnull) inputMatrix(MPSVector
*__nonnull) meanVector(MPSVector *__nonnull)
varianceVector(MPSVector *__nullable) gammaVector(MPSVector
*__nullable) betaVector(MPSMatrix *__nonnull)
resultGradientForDataMatrix(MPSVector *__nullable)
resultGradientForGammaVector(MPSVector *__nullable)
resultGradientForBetaVector
Encode a MPSMatrixBatchNormalizationGradient object to a
command buffer and compute its gradient with respect to its input data.
Parameters:
commandBuffer The commandBuffer on which to encode
the operation.
gradientMatrix A matrix whose values represent the gradient of a
loss function with respect to the results of a forward
MPSMatrixBatchNormalization operation.
inputMatrix A matrix containing the inputs to a forward
MPSMatrixBatchNormalization operation for which the gradient values are
to be computed.
meanVector A vector containing the batch mean values. Should
contain either the specified values used to compute the forward result, or the
computed values resulting from the forward kernel execution.
varianceVector A vector containing the batch variance values.
Should contain either the specified values used to compute the forward result,
or the computed values resulting from the forward kernel execution.
gammaVector A vector containing the gamma terms. Should be the
same values as used when computing the forward result.
betaVector A vector containing the beta terms. Should be the same
values as used when computing the forward result.
resultGradientForDataMatrix The matrix containing the resulting gradient
values.
resultGradientForGammaVector If non-NULL the vector containing gradients
for the gamma terms.
resultGradientForBetaVector If non-NULL the vector containing gradients
for the beta terms.
- (nullable instancetype) initWithCoder: (NSCoder
*__nonnull) aDecoder(nonnull id< MTLDevice >) device
NSSecureCoding compatability See
MPSKernel::initWithCoder.
Parameters:
aDecoder The NSCoder subclass with your serialized
MPSMatrixBatchNormalizationGradient
device The MTLDevice on which to make the
MPSMatrixBatchNormalizationGradient object.
Returns:
A new MPSMatrixBatchNormalizationGradient
object, or nil if failure.
Reimplemented from MPSKernel.
- (nonnull instancetype) initWithDevice: (nonnull id<
MTLDevice >) device
Standard init with default properties per filter type
Parameters:
device The device that the filter will be used on.
May not be NULL.
Returns:
a pointer to the newly initialized object. This will
fail, returning nil if the device is not supported. Devices must be
MTLFeatureSet_iOS_GPUFamily2_v1 or later.
Reimplemented from MPSKernel.
- (float) neuronParameterA
Getter funtion for neuronType set using
setNeuronType:parameterA:parameterB:parameterC method
- (float) neuronParameterB
Getter funtion for neuronType set using
setNeuronType:parameterA:parameterB:parameterC method
- (float) neuronParameterC
Getter funtion for neuronType set using
setNeuronType:parameterA:parameterB:parameterC method
- (MPSCNNNeuronType) neuronType
Getter funtion for neuronType set using
setNeuronType:parameterA:parameterB:parameterC method
- (void) setNeuronType: (MPSCNNNeuronType)
neuronType(float) parameterA(float) parameterB(float) parameterC
Specifies a neuron activation function to be used.
This method can be used to add a neuron activation funtion of
given type with associated scalar parameters A, B, and C that are
shared across all output values. Note that this method can only be used to
specify neurons which are specified by three (or fewer) parameters shared
across all output values (or channels, in CNN nomenclature). It is an error
to call this method for neuron activation functions like
MPSCNNNeuronTypePReLU, which require per-channel parameter values. An
MPSMatrixBatchNormalizationGradient kernel is initialized with a
default neuron function of MPSCNNNeuronTypeNone.
Parameters:
neuronType Type of neuron activation function. For
full list see MPSCNNNeuronType.h
parameterA parameterA of neuron activation that is shared across all
output values.
parameterB parameterB of neuron activation that is shared across all
output values.
parameterC parameterC of neuron activation that is shared across all
output values.
- epsilon [read], [write], [nonatomic],
[assign]
A small term added to the variance when normalizing the
input.
- sourceInputFeatureChannels [read], [write],
[nonatomic], [assign]
The number of feature channels in the input vectors.
- sourceNumberOfFeatureVectors [read], [write],
[nonatomic], [assign]
The number of input vectors which make up the input array.
Generated automatically by Doxygen for
MetalPerformanceShaders.framework from the source code.