Batch Processing

Layer Input

The backward two-dimensional convolution layer accepts the input described below. Pass the Input ID as a parameter to the methods that provide input for your algorithm. For more details, see Algorithms.

Input ID

Input

inputGradient

Pointer to tensor G of size n1 x nKernels x l3 x l4 that stores the input gradient computed on the preceding layer. This input can be an object of any class derived from Tensor.

inputFromForward

Collection of input data needed for the backward two-dimensional convolution layer. This collection can contain objects of any class derived from Tensor.

Element ID

Element

auxData

Pointer to tensor X of size n1 × n2 × n3 × n4 that stores the input data for the forward two-dimensional convolution layer. This input can be an object of any class derived from Tensor.

auxWeights

Pointer to tensor K of size nKernels × m2 × m3 × m4 that stores a set of kernel weights. This input can be an object of any class derived from Tensor.

Layer Parameters

For common parameters of neural network layers, see Common Parameters.

In addition to the common parameters, the backward two-dimensional convolution layer has the following parameters:

Parameter

Default Value

Description

algorithmFPType

float

The floating-point type that the algorithm uses for intermediate computations. Can be float or double.

method

defaultDense

Performance-oriented computation method, the only method supported by the layer.

kernelSizes

KernelSizes(2, 2)

Data structure representing the sizes mi, i ∈ {3, 4}, of the two-dimensional kernel subtensor.

indices

Indices(2,3)

Data structure representing the dimensions for applying convolution kernels.

strides

Strides(2, 2)

Data structure representing the intervals si, i ∈ {3, 4}, on which the kernel should be applied to the input.

paddings

Paddings(0, 0)

Data structure representing the number of data elements pi, i ∈ {3, 4}, to implicitly add to each side of the two-dimensional subtensor along which forward two-dimensional convolution is performed.

nKernels

n/a

Number of kernels applied to the input layer data.

groupDimension

1

Dimension for which grouping is applied.

nGroups

1

Number of groups into which the input data is split in dimension groupDimension.

propagateGradient

false

Flag that specifies whether the backward layer propagates the gradient.

Layer Output

The backward two-dimensional convolution layer calculates the result described below. Pass the Result ID as a parameter to the methods that access the results of your algorithm. For more details, see Algorithms.

Result ID

Result

gradient

Pointer to tensor Z of size n1 × n2 × n3 × n4 that stores the result of the backward two-dimensional convolution layer. This result can be an object of any class derived from Tensor.

weightDerivatives

Pointer to the tensor of size nKernels × m2 × m3 × m4 that stores result ∂Ε /krcuv of the backward two-dimensional convolution layer, where r = {0, ..., nKernels - 1}, c = {0, ..., m2 - 1}, u = {0, ..., m3 - 1}, v = {0, ..., m4 - 1}. This result can be an object of any class derived from Tensor.

biasDerivatives

Pointer to the tensor of size nKernels that stores result ∂Ε / ∂br of the backward two-dimensional convolution layer, where r = {0, ..., nKernels - 1}. This result can be an object of any class derived from Tensor.

For more complete information about compiler optimizations, see our Optimization Notice.
Select sticky button color: 
Orange (only for download buttons)