org.neuroph.nnet.comp.layer

Class PoolingLayer

java.lang.Object

org.neuroph.core.Layer

org.neuroph.nnet.comp.layer.FeatureMapsLayer

org.neuroph.nnet.comp.layer.PoolingLayer

All Implemented Interfaces:

Serializable, Iterable<Neuron>

public class PoolingLayer
extends FeatureMapsLayer

Pooling layer is a special type of feature maps layer (FeatureMapsLayer) which is used in convolutional networks. It contains neurons with max input function and method for creating pooling layer specific conectivity patterns. The role of pooling layer is dimensionality and complexity reduction, while it keeps essential information.

Author:

Boris Fulurija, Zoran Sevarac

See Also:

FeatureMapsLayer, Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
static NeuronProperties	DEFAULT_NEURON_PROP Default neuron properties for pooling layer

Fields inherited from class org.neuroph.nnet.comp.layer.FeatureMapsLayer

mapDimensions

Fields inherited from class org.neuroph.core.Layer

neurons

Constructor Summary

Constructors

Constructor and Description
PoolingLayer(FeatureMapsLayer fromLayer, Dimension2D kernelDim) Creates pooling layer with specified kernel, appropriate map dimensions in regard to previous layer (fromLayer param) and specified number of feature maps with default neuron settings for pooling layer.
PoolingLayer(FeatureMapsLayer fromLayer, Dimension2D kernelDim, int numberOfMaps, NeuronProperties neuronProp) Creates pooling layer with specified kernel, appropriate map dimensions in regard to previous layer (fromLayer param) and specified number of feature maps with given neuron properties.

Method Summary

Modifier and Type	Method and Description
void	connectMaps(FeatureMapLayer fromMap, FeatureMapLayer toMap) Creates connections with shared weights between two feature maps Assumes that toMap is from Pooling layer.

Methods inherited from class org.neuroph.nnet.comp.layer.FeatureMapsLayer

addFeatureMap, createFeatureMaps, getFeatureMap, getFeatureMaps, getMapDimensions, getNeuronAt, getNeuronsCount, getNumberOfMaps

Methods inherited from class org.neuroph.core.Layer

addNeuron, addNeuron, calculate, getLabel, getNeuronAt, getNeurons, getParentNetwork, indexOf, initializeWeights, isEmpty, iterator, removeAllNeurons, removeNeuron, removeNeuronAt, reset, setLabel, setNeuron, setParentNetwork

Methods inherited from class java.lang.Object

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

Methods inherited from interface java.lang.Iterable

forEach, spliterator

Field Detail

DEFAULT_NEURON_PROP

public static final NeuronProperties DEFAULT_NEURON_PROP

Default neuron properties for pooling layer

Constructor Detail

PoolingLayer

public PoolingLayer(FeatureMapsLayer fromLayer,
                    Dimension2D kernelDim)

Creates pooling layer with specified kernel, appropriate map dimensions in regard to previous layer (fromLayer param) and specified number of feature maps with default neuron settings for pooling layer. Number of maps in pooling layer must be the same as number of maps in previous layer.

Parameters:

fromLayer - previous layer, which will be connected to this layer

kernel - kernel for all feature maps

PoolingLayer

public PoolingLayer(FeatureMapsLayer fromLayer,
                    Dimension2D kernelDim,
                    int numberOfMaps,
                    NeuronProperties neuronProp)

Creates pooling layer with specified kernel, appropriate map dimensions in regard to previous layer (fromLayer param) and specified number of feature maps with given neuron properties.

Parameters:

fromLayer - previous layer, which will be connected to this layer

kernel - kernel for all feature maps

numberOfMaps - number of feature maps to create in this layer

neuronProp - settings for neurons in feature maps

Method Detail

connectMaps

public void connectMaps(FeatureMapLayer fromMap,
                        FeatureMapLayer toMap)

Creates connections with shared weights between two feature maps Assumes that toMap is from Pooling layer.

In this implementation, there is no overlapping between kernel positions.

Specified by:

connectMaps in class FeatureMapsLayer

Parameters:

fromMap - source feature map

toMap - destination feature map