Developer Guide for Intel® Data Analytics Acceleration Library 2019 Update 5
The backward fully-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 the tensor of size nk x m 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 data needed for the backward fully-connected layer. This collection can contain objects of any class derived from Tensor. |
|
Element ID |
Element |
|
auxData |
Pointer to the tensor of size n1 x ... x nk x ... x np that stores the input data from the forward fully-connected layer. This input can be an object of any class derived from Tensor. |
|
auxWeights |
Pointer to the tensor of size n1 x ... x nk - 1 x m x nk + 1 x ... x np that stores a set of weights. This input can be an object of any class derived from Tensor. |
|
auxBiases |
Pointer to the tensor of size m that stores a set of biases. This input can be an object of any class derived from Tensor. |
|
auxMask |
Pointer to the tensor of size n1 x ... x nk - 1 x m x nk + 1 x ... x np that holds 1 for the corresponding weights used in computations and 0 for the weights not used in computations. If no mask is provided, the library uses all the weights. |
For common parameters of neural network layers, see Common Parameters.
In addition to the common parameters, the backward fully-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. |
|
nOutputs |
Not applicable |
Number of layer outputs m. Required to initialize the algorithm. |
|
propagateGradient |
false |
Flag that specifies whether the backward layer propagates the gradient. |
The backward fully-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 the tensor of size n1 x ... x nk x ... x np that stores the result of the backward fully-connected layer. This input can be an object of any class derived from Tensor. |
|
weightDerivatives |
Pointer to the tensor of size n1 x ... x nk - 1 x m x nk + 1 x ... x np that stores result ∂Ε / ∂wi1...j...ip of the backward fully-connected layer, where j = {1, ..., m}. This input can be an object of any class derived from Tensor. |
|
biasDerivatives |
Pointer to the tensor of size m that stores result ∂Ε / ∂bj of the backward fully-connected layer, where j = {1, ..., m}. This input can be an object of any class derived from Tensor. |
C++: fullycon_layer_dense_batch.cpp
Java*: FullyconLayerDenseBatch.java
Python*: fullycon_layer_dense_batch.py