---

gnn_affine : Affine Activation Function.
[Atomic Nodes.]

---

Detailed Description

This node type implements an Affine activatino function given by:

where the index runs over all inputs / outputs. The amplitude factors and the biases are independent and can be trained. Usually, this activation function is placed after a non-linear activation function which is limited in its output range, like a gaussian, logistic sigmoid or a hyperbolic tangent activation function, to scale the outputs to the correct magnitudes.

The function's gradients are:

with respect to its inputs, and

with respect to its parameters.

Typedefs
typedef _gnn_affine	gnn_affine
	The structure for a gnn_affine : Affine Activation Function. node.
Functions
int	gnn_affine_f (gnn_node *node, const gsl_vector *x, const gsl_vector *w, gsl_vector *y)
	Computes the output.
int	gnn_affine_dx (gnn_node *node, const gsl_vector *x, const gsl_vector *w, const gsl_vector *dy, gsl_vector *dx)
	Computes .
int	gnn_affine_dw (gnn_node *node, const gsl_vector *x, const gsl_vector *w, const gsl_vector *dy, gsl_vector *dw)
	Computes .
void	gnn_affine_destroy (gnn_node *node)
	Destructor function.
gnn_node *	gnn_affine_new (int input_size, double a, double b)
	Creates an Affine Activation Function node.
gnn_node *	gnn_affine_standard_new (int input_size)
	Creates a standard Affine Activation Function node.

---

Typedef Documentation

typedef struct _gnn_affine gnn_affine

This datatype holds the information for a gnn_affine : Affine Activation Function. node. Basically, it extends the gnn_node with special pointers to get fast accesses to the amplitudes and variances. Definition at line 48 of file gnn_affine.h.

---

Function Documentation

void gnn_affine_destroy (  gnn_node *    node )  [static]

Parameters: node  A pointer to the gnn_affine : Affine Activation Function. node. Definition at line 205 of file gnn_affine.c.

int gnn_affine_dw (  gnn_node *    node, const gsl_vector *    x, const gsl_vector *    w, const gsl_vector *    dy, gsl_vector *    dw )  [static]

Parameters: node  A pointer to the gnn_affine : Affine Activation Function. node. x  The input vector . w  The parameter vector . dy  The vector . dx  The output buffer vector . Returns: Returns 0 on success. Definition at line 172 of file gnn_affine.c.

int gnn_affine_dx (  gnn_node *    node, const gsl_vector *    x, const gsl_vector *    w, const gsl_vector *    dy, gsl_vector *    dx )  [static]

Parameters: node  A pointer to the gnn_affine : Affine Activation Function. node. x  The input vector . w  The parameter vector . dy  The vector . dx  The output buffer vector . Returns: Returns 0 on success. Definition at line 137 of file gnn_affine.c.

int gnn_affine_f (  gnn_node *    node, const gsl_vector *    x, const gsl_vector *    w, gsl_vector *    y )  [static]

Parameters: node  A pointer to the gnn_affine : Affine Activation Function. node. x  The input vector . w  The parameter vector . y  The output buffer vector . Returns: Returns 0 on success. Definition at line 102 of file gnn_affine.c.

gnn_node* gnn_affine_new (  int    input_size, double    a, double    b )

This function creates a node of the gnn_affine type. Although the parameters and , are all equal initially, they change during training. Parameters: input_size  The input size . a  The (amplitude) factors. v  The biases . Returns: A pointer to a new gnn_divergence : Input Divergence. node. Definition at line 235 of file gnn_affine.c.

gnn_node* gnn_affine_standard_new (  int    input_size )

This function creates a node of the gnn_affine type. The parameters are initialized with amplitudes and biases . Parameters: input_size  The input size . Returns: A pointer to a new gnn_tanh node. Definition at line 340 of file gnn_affine.c.

---