boone.neurons

Class MaxPoolNeuron

java.lang.Object

boone.BrainPart

boone.Neuron

boone.neurons.MaxPoolNeuron

All Implemented Interfaces:

Storable, java.lang.Cloneable, java.lang.Comparable

public class MaxPoolNeuron
extends Neuron
implements Storable, java.lang.Cloneable, java.lang.Comparable

A neuron for a max pooling layer. It does not sum up the input values, but sets the maximal input value to be the global link input.

Since:

July 2021

Version:

$Id$

Author:

Helmut A. Mayer

Field Summary

Fields inherited from class boone.Neuron

errorSignal, externalInput, input, inputNeuron, layer, linkInput, links, output, outputNeuron, position, usingBias

Fields inherited from class boone.BrainPart

activationFn, name, NO_ID, weight

Constructor Summary

Constructors

Constructor and Description
MaxPoolNeuron() Creates a default MaxPoolNeuron with identity activation function and no bias use.

Method Summary

Modifier and Type	Method and Description
void	addLinkInput(Link link) Registers the maximal input value to be used as global link input.
void	calculate() Calculates the neuron output based on its input, which is the maximal signal.
MaxPoolNeuron	clone() Returns a semi-deep clone of this neuron.

Methods inherited from class boone.Neuron

addToBias, compareTo, fromXML, getBias, getErrorSignal, getExternalInput, getInput, getInputLink, getInputLinkCount, getLayer, getLink, getLinkCount, getLinkInput, getLinks, getLinkTo, getOutput, getOutputLink, getOutputLinkCount, getPosition, isHiddenNeuron, isInputNeuron, isOutputNeuron, isUsingBias, propagateOutput, reset, setBias, setErrorSignal, setExternalInput, setInput, setInputNeuron, setLayer, setLinkInput, setOutput, setOutputNeuron, setPosition, setUsingBias, toString, toXML

Methods inherited from class boone.BrainPart

addGradient, addToWeight, getAccu, getActivationFn, getGradient, getID, getLastGradient, getName, getUniqueID, getWeight, isTrainable, randomize, randomize, setAccu, setActivationFn, setGradient, setID, setLastGradient, setName, setTrainable, setWeight

Methods inherited from class java.lang.Object

equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface boone.io.Storable

fromXML, toXML

Methods inherited from interface java.lang.Comparable

compareTo

Constructor Detail

MaxPoolNeuron

public MaxPoolNeuron()

Creates a default MaxPoolNeuron with identity activation function and no bias use.

Method Detail

clone

public MaxPoolNeuron clone()

Returns a semi-deep clone of this neuron. The clone's link list is empty.

Overrides:

clone in class Neuron

Returns:

a shallow clone

addLinkInput

public void addLinkInput(Link link)

Registers the maximal input value to be used as global link input. Also, sets the input weights such that the weights reflect max pooling, i.e., the weight of the link with maximal input value is set to 1.0, the others are set to 0.0. This is needed for correct error back-propagation, although the pooling links are not trained.

Overrides:

addLinkInput in class Neuron

Parameters:

link - a link sending a value/signal

calculate

public void calculate()

Calculates the neuron output based on its input, which is the maximal signal.

Overrides:

calculate in class Neuron