org.deeplearning4j.nn.conf.weightnoise

Class DropConnect

java.lang.Object

org.deeplearning4j.nn.conf.weightnoise.DropConnect

All Implemented Interfaces:

Serializable, Cloneable, IWeightNoise

public class DropConnect
extends Object
implements IWeightNoise

DropConnect, based on Wan et. al 2013 - "Regularization of Neural Networks using DropConnect"
Sets weights randomly to 0 with some probability, or leaves them unchanged.

Author:

Alex Black

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
DropConnect(double weightRetainProbability)
DropConnect(double weightRetainProbability, boolean applyToBiases)
DropConnect(org.nd4j.linalg.schedule.ISchedule weightRetainProbSchedule)
DropConnect(org.nd4j.linalg.schedule.ISchedule weightRetainProbSchedule, boolean applyToBiases)

Method Summary

Modifier and Type	Method and Description
DropConnect	clone()
org.nd4j.linalg.api.ndarray.INDArray	getParameter(Layer layer, String paramKey, int iteration, int epoch, boolean train, LayerWorkspaceMgr workspaceMgr) Get the parameter, after applying weight noise

Methods inherited from class java.lang.Object

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

Constructor Detail

DropConnect

public DropConnect(double weightRetainProbability)

Parameters:

weightRetainProbability - Probability of retaining a weight

DropConnect

public DropConnect(double weightRetainProbability,
                   boolean applyToBiases)

Parameters:

weightRetainProbability - Probability of retaining a weight

applyToBiases - If true: apply to biases (default: weights only)

DropConnect

public DropConnect(org.nd4j.linalg.schedule.ISchedule weightRetainProbSchedule)

Parameters:

weightRetainProbSchedule - Probability (schedule) of retaining a weight

DropConnect

public DropConnect(org.nd4j.linalg.schedule.ISchedule weightRetainProbSchedule,
                   boolean applyToBiases)

Parameters:

weightRetainProbSchedule - Probability (schedule) of retaining a weight

applyToBiases - If true: apply to biases (default: weights only)

Method Detail

getParameter

public org.nd4j.linalg.api.ndarray.INDArray getParameter(Layer layer,
                                                         String paramKey,
                                                         int iteration,
                                                         int epoch,
                                                         boolean train,
                                                         LayerWorkspaceMgr workspaceMgr)

Description copied from interface: IWeightNoise

Get the parameter, after applying weight noise

Specified by:

getParameter in interface IWeightNoise

Parameters:

layer - Layer to get the parameter for

paramKey - Parameter key

iteration - Iteration number

epoch - Epoch number

train - If true: training. False: at test time

Returns:

Parameter, after applying weight noise

clone

public DropConnect clone()

Specified by:

clone in interface IWeightNoise

Overrides:

clone in class Object