Developer Guide for Intel® Data Analytics Acceleration Library 2019 Update 4
The backward two-dimensional locally-connected 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 locally-connected 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 locally-connected layer. This input can be an object of any class derived from Tensor. |
|
auxWeights |
Pointer to tensor K of size nKernels × l3 × l4 × m2 × m3 × m4 that stores a set of kernel weights. This input can be an object of any class derived from Tensor. |
For common parameters of neural network layers, see Common Parameters.
In addition to the common parameters, the backward two-dimensional locally-connected 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 kernels of the backward locally-connected layer. |
|
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 to which the kernel are applied. Only symmetric padding is currently supported. |
|
nKernels |
n/a |
Number of kernels applied to the input layer data. |
|
groupDimension |
1 |
Dimension n2 for which grouping is applied. Only groupDimension = 1 is currently supported. |
|
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. |
The backward two-dimensional locally-connected 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 locally-connected layer. This input can be an object of any class derived from Tensor. |
|
weightDerivatives |
Pointer to the tensor of size nKernels × l3 × l4 × m2 × m3 × m4 that stores result ∂Ε / ∂krijcuv of the backward two-dimensional locally-connected layer. This input can be an object of any class derived from Tensor. |
|
biasDerivatives |
Pointer to the tensor of size nKernels × l3 × l4 that stores result ∂Ε / ∂brij of the backward two-dimensional locally-connected layer. This input can be an object of any class derived from Tensor. |
C++: locallycon2d_layer_dense_batch.cpp
Java*: Locallycon2DLayerDenseBatch.java
Python*: locallycon2d_layer_dense_batch.py