is a pair of input & output, such as 2 -> 3
is an activation function to be applied
is initial weight matrix for the case that it is restored from JSON (default: null)
is inital bias matrix for the case that it is restored from JSON (default: null)
is initial reconstruct bias matrix for the case that it is restored from JSON (default: null)
weights for update
an activation function to be applied
an activation function to be applied
Forward computation
Forward computation
input matrix
output matrix
accumulated delta values
Sugar: reconstruction
Sugar: reconstruction
hidden layer output matrix
tuple of reconstruction output
Backpropagation of reconstruction.
Backpropagation of reconstruction. For the information about backpropagation calculation, see kr.ac.kaist.ir.deep.layer.Layer
error matrix to be propagated
input of this layer
final reconstruction output of this layer
propagated error
Number of Fan-ins
Number of Fan-ins
Number of output
Number of output
Sugar: Forward computation.
Sugar: Forward computation. Calls apply(x)
input matrix
output matrix
Translate this layer into JSON object (in Play! framework)
Translate this layer into JSON object (in Play! framework)
JSON object describes this layer
Backward computation.
Backward computation.
to be propagated ( dG / dF
is propagated from higher layer )
of this layer (in this case, x = entry of dX / dw
)
of this layer (in this case, y
)
propagated error (in this case, dG/dx
)
Let this layer have function F composed with function X(x) = W.x + b
and higher layer have function G.
Weight is updated with: dG/dW
and propagate dG/dx
For the computation, we only used denominator layout. (cf. Wikipedia Page of Matrix Computation) For the computation rules, see "Matrix Cookbook" from MIT.
Layer : Reconstructable Basic Layer