---

fl::NeuronBackprop Class Reference

#include <neural.h>

Inheritance diagram for fl::NeuronBackprop:

List of all members.

Public Methods
	NeuronBackprop ()
virtual void	startCycle ()
	Clear latches.
float	getActivation ()
	Add up net activation level from all synapses.
virtual float	getOutput ()
	Compute output of squashing function. Draws on net activation.
virtual float	getDelta ()
	Collect error from subsequent Neurons via back-propogation.
float	getError ()
	Compute the error signal based on delta.
virtual void	learn ()
	Request all Synapses feeding this Neuron to adjust their weights according to the error signal.
Public Attributes
float	activation
	net activation level (before squashing function) for current cycle. Before it is calculated, its value is nan.
float	delta
	error signal for current cycle. Before it is calculated, its value is nan.

---

Detailed Description

Computes the neuron's response by summing inputs and passing thru a squashing function. Performs back-propogation learning. Works in a rigid one-shot fashion, so must be reset and reloaded to handle mutliple data or points in time.

To construct an NN that works on real world data and has useful output, you should create two subclasses of this class: "NeuronInput" -- Computes an appropriate output level based on current input datum. Only need to override getOutput(). No need for afferent Synapses. "NeuronOutput" -- Computes an appropriate error signal based on the expected value from a training datum. Only need to override getDelta().

Note that hidden layers in an NN would be constructed directly out of instances of this class or some specialized subclass. An example of specialization is the NeuronDelay below.

---

Constructor & Destructor Documentation

NeuronBackprop::NeuronBackprop (    )

---

Member Function Documentation

float NeuronBackprop::getActivation (    )

Add up net activation level from all synapses.

float NeuronBackprop::getDelta (    )  [virtual]

Collect error from subsequent Neurons via back-propogation. Reimplemented in fl::NeuronDelay, fl::NeuronOutputVector, and NeuronOutput.

float NeuronBackprop::getError (    )

Compute the error signal based on delta.

float NeuronBackprop::getOutput (    )  [virtual]

Compute output of squashing function. Draws on net activation. Reimplemented in fl::NeuronDelay, fl::NeuronInputVector, and NeuronInput.

void NeuronBackprop::learn (    )  [virtual]

Request all Synapses feeding this Neuron to adjust their weights according to the error signal.

void NeuronBackprop::startCycle (    )  [virtual]

Clear latches. Reimplemented in fl::NeuronDelay.

---

Member Data Documentation

float fl::NeuronBackprop::activation

net activation level (before squashing function) for current cycle. Before it is calculated, its value is nan.

float fl::NeuronBackprop::delta

error signal for current cycle. Before it is calculated, its value is nan.

---

The documentation for this class was generated from the following files:

• neural.h
• NeuralNetworkBackprop.cc

---