Returns the gradient (Layer-valued) and the backpropagated error, which is passed to the layer below.
Returns the gradient (Layer-valued) and the backpropagated error, which is passed to the layer below.
This method can rely on the fact that the propagate method
already has been called in the first pass.
error propagated from above, formatted the same way (one row for each example) as input and output
gradient (Layer-valued) and the next backpropagated error
Returns the output given the input.
Returns the output given the input. This method can cache data that could be useful on the second pass of the backpropagation.
The input contains one example in each row, the output shall have the same layout.
Creates a new independent layer that has the same type as this one, but propagates the information in reverse direction
Creates a new independent layer that has the same type as this one, but propagates the information in reverse direction
Returns the result of signal propagation in reverse direction
Returns the result of signal propagation in reverse direction
Color map for the activities
Color map for the activities
Optionally, one can specify how to reshape the neuron activities for visualization (height, width).
Optionally, one can specify how to reshape the neuron activities for visualization (height, width).