Class ConvolutionalNetwork

java.lang.Object

deepnetts.net.NeuralNetwork<BackpropagationTrainer>

deepnetts.net.ConvolutionalNetwork

All Implemented Interfaces:

TrainerProvider<BackpropagationTrainer>, Serializable, AutoCloseable

public class ConvolutionalNetwork
extends NeuralNetwork<BackpropagationTrainer>
implements Serializable

Convolutional neural network is an extension of feed forward network, which can include 2D and 3D adaptive preprocessing layers (Convolutional and MaxPooling layer), which is specialized to learn to recognize features in images. Images are fed as 3-dimensional tensors (multidimensional arrays). Although primary used for images, they can also be applied to other types of problems.

Author:

Zoran Sevarac

See Also:

• ConvolutionalLayer
• MaxPoolingLayer
• BackpropagationTrainer
• Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	ConvolutionalNetwork.Builder	Builder for a convolutional neural network.

Field Summary

Fields inherited from class NeuralNetwork

cudaHandles, threadPool

Method Summary

Modifier and Type	Method	Description
static ConvolutionalNetwork.Builder	builder()	Returns a builder for the ConvolutionalNetwork
List<TensorBase>	getDeltaWeights()	Returns delta weights for all layers.
List<TensorBase>	getLayersGradients()
List<TensorBase>	getLayersOutputs()	Returns outputs of all layers.
List<TensorBase>	getWeights()	Returns weights from all layers in this network as a list of tensors.
void	setInput(TensorBase input)	Sets network input and calculates entire network (triggers forward pass).
void	setWeights(List<String> weights)	Sets network's weights for all layers.

Methods inherited from class NeuralNetwork

addLayer, applyWeightChanges, backward, close, forward, getInputLayer, getL1RegSum, getL2RegSum, getLabel, getLayerAt, getLayers, getLossFunction, getMode, getNormalizer, getOutput, getOutputAsTensor, getOutputLabel, getOutputLabels, getOutputLayer, getPreprocessing, getThreadPool, getTrainer, load, predict, save, setInput, setInputLayer, setLabel, setLossFunction, setMode, setNormalizer, setOutputError, setOutputLabels, setOutputLayer, setPreprocessing, setTrainer, test, toString, train

Methods inherited from class Object

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

Method Details

setInput

public void setInput(TensorBase input)

Description copied from class: NeuralNetwork

Sets network input and calculates entire network (triggers forward pass).

Overrides:

setInput in class NeuralNetwork<BackpropagationTrainer>

Parameters:

input - input tensor

builder

public static ConvolutionalNetwork.Builder builder()

Returns a builder for the ConvolutionalNetwork

Returns:

builder instance

getWeights

public List<TensorBase> getWeights()

Returns weights from all layers in this network as a list of tensors.

Returns:

all network's weights

setWeights

public void setWeights(List<String> weights)

Sets network's weights for all layers.

Parameters:

weights - List of weights for all layers.

getDeltaWeights

public List<TensorBase> getDeltaWeights()

Returns delta weights for all layers. Delta weights are weight changes calculated during the training procedure. Useful for debugging

Returns:

getLayersOutputs

public List<TensorBase> getLayersOutputs()

Returns outputs of all layers. Useful for debugging

Returns:

getLayersGradients

public List<TensorBase> getLayersGradients()