where 
with 
are matrices of rank 
associated with the delays 
. The vectors 
with
area the input vectors to the node. It is important to note that the real input 
is equal to 
, and that the vectors 
are the input vectors
presented to the node in the previous 
iterations, which have been buffered by the node.
In other words, each output 
, 
consists of
where the 
can be seen as casual discrete filters of order
, and so 
are the convolutions of the signal
with the FIR (finite impulse response) filters
.
The corresponding gradients are:
or, written in matrix notation:
and
or:
Typedefs | |
| typedef _gnn_fir | gnn_fir |
| The structure for a gnn_fir : FIR Synaptic Filters node. | |
Functions | |
| int | gnn_fir_f (gnn_node *node, const gsl_vector *x, const gsl_vector *w, gsl_vector *y) |
| Computes the evaluation. | |
| int | gnn_fir_dx (gnn_node *node, const gsl_vector *x, const gsl_vector *w, const gsl_vector *dy, gsl_vector *dx) |
| Computes the gnn_fir : FIR Synaptic Filters gradient with respect to its inputs. | |
| int | gnn_fir_dw (gnn_node *node, const gsl_vector *x, const gsl_vector *w, const gsl_vector *dy, gsl_vector *dw) |
| Computes the gnn_fir : FIR Synaptic Filters gradient with respect to its parameters. | |
| void | gnn_fir_destroy (gnn_node *node) |
| The destroy function for a gnn_fir : FIR Synaptic Filters. | |
| gnn_node * | gnn_fir_new (size_t input_size, size_t output_size, size_t delay) |
| Creates a new gnn_fir : FIR Synaptic Filters node. | |
| int | gnn_fir_init (gnn_node *node) |
| Initializes the node's parameters. | |
|
|
This datatype holds the information for a gnn_fir : FIR Synaptic Filters node. Basically, it extends the gnn_node with special pointers to get fast accesses to the matrices These views exist for the raw parameters, their derivatives and their frozen flags, in order to improve the efficiency. |
|
|
This functions is the gnn_fir : FIR Synaptic Filters destroy function, which destroys the type's specific data.
|
|
||||||||||||||||||||||||
|
|
|
||||||||||||||||||||||||
|
|
|
||||||||||||||||||||
|
|
|
|
This functions initializes the layer's parameters by values drawn from a uniform distribution in
|
|
||||||||||||||||
|
This function creates a new gnn_node of gnn_fir : FIR Synaptic Filters type.
|
1.2.18
and to the buffered input vector 
.
. 
. 
where the result should be placed. 
where the result should be placed. 
where the result should be placed. 
.