replaceLayer

Replace layer in neural network

Syntax

netUpdated = replaceLayer(net,layerName,layers)

netUpdated = replaceLayer(net,layerPath,layers)

netUpdated = replaceLayer(___,ReconnectBy=mode)

Description

netUpdated = replaceLayer(net,layerName,layers) replaces the layer layerName in the dlnetwork object net with the layers in layers.

replaceLayer connects the layers in layers sequentially and connects layers into the network.

netUpdated = replaceLayer(net,layerPath,layers) replaces the layer specified by layerPath in the dlnetwork object net with the layers in layers. (since R2024a)

netUpdated = replaceLayer(___,ReconnectBy=mode) additionally specifies the method of reconnecting layers.

Examples

Replace Layer in Neural Network

Define a simple network architecture and display the network in a plot.

net = dlnetwork;

layers = [
    imageInputLayer([28 28 1],Name="input")
    convolution2dLayer(3,16,Padding="same")
    reluLayer(Name="relu1")
    additionLayer(2,Name="add")
    fullyConnectedLayer(10)
    softmaxLayer];

net = addLayers(net,layers);
net = connectLayers(net,"input","add/in2");

figure
plot(net)

Replace the ReLU layer in the network with a batch normalization layer followed by a leaky ReLU layer.

layers = [
    batchNormalizationLayer
    leakyReluLayer(0.1)];

net = replaceLayer(net,"relu1",layers);

Plot the updated network.

figure
plot(net)

Replace Layer in Nested Network

Define a network architecture containing several network layers.

net = dlnetwork;

layers = [lstmLayer(100,OutputMode="sequence")
    dropoutLayer(0.2)];

lstmDropoutLayer = networkLayer(layers);

layers = [sequenceInputLayer(3)
    lstmDropoutLayer
    lstmDropoutLayer
    fullyConnectedLayer(10)
    softmaxLayer];

net = addLayers(net,layers);

Plot the network.

figure
plot(net)

Look up the first LSTM layer by specifying the path to the layer. The path includes the name of the network layer ("subnet_1") and the name of the LSTM layer ("lstm"), separated by a forward slash. You can use the path to look up layers nested within a hierarchy of multiple network layers.

tempLSTMLayer = getLayer(net,"subnet_1/lstm");

Edit properties of the layer, and replace the original layer with the modified layer.

tempLSTMLayer.InputWeightsInitializer = "zeros";
tempLSTMLayer.RecurrentWeightsInitializer = "zeros";
tempLSTMLayer.BiasInitializer = "ones";
tempLSTMLayer.Name = "modifiedLSTM";

Replace the original layer with the modified layer, specifying the network, the path to the layer to replace, and the modified layer.

net = replaceLayer(net,"subnet_1/lstm",tempLSTMLayer);

Input Arguments

net — Neural network

dlnetwork object

Neural network, specified as a dlnetwork object.

layerName — Name of layer to replace

string scalar | character vector

Name of the layer to replace, specified as a string scalar or a character vector.

layers — Network layers

Layer array

Network layers, specified as a Layer array.

For a list of built-in layers, see List of Deep Learning Layers.

layerPath — Path to nested layer

character vector | string scalar

Path to nested layer, specified as a character vector or string scalar.

For a layer within a networkLayer, specify layerPath as:

• The name of the network layer and the name of the nested layer separated by a forward slash "/". For example, the path to a layer named "nestedLayerName" in a network layer named "networkLayerName" is "networkLayerName/nestedLayerName". If there are multiple levels of nested layers, then specify the path using the form networkLayerName1/.../networkLayerNameN/nestedLayerName.

Custom layers can have a dlnetwork as a learnable or state property. For a layer of a dlnetwork in the property of a custom layer, specify layerPath as:

• The name of the custom layer, the name of the dlnetwork property, and the name of the nested layer separated by forward slashes "/". For example, the path to a layer named "layerName" inside a dlnetwork property named "propertyName" of a custom layer named "customLayerName" is "customLayerName/propertyName/layerName". If there are multiple levels of nested layers, then specify the path using the form "customLayerName1/propertyName1/.../customLayerNameN/propertyNameN/layerName".

Data Types: char | string

mode — Method to reconnect layers

"name" (default) | "order"

Method to reconnect layers specified as one of the following:

• "name" – Reconnect layers using the input and output names of the replaced layer. For each layer connected to an input of the replaced layer, reconnect the layer to the input of the same input name of layers(1). For each layer connected to an output of the replaced layer, reconnect the layer to the output of the same output name of layers(end).

• "order" – Reconnect layers using the order of the input names of layers(1) and the output names of layers(end). Reconnect the layer connected to the ith input of the replaced layer to the ith input of layers(1). Reconnect the layer connected to the jth output of the replaced layer to the jth output of layers(end).

Output Arguments

netUpdated — Updated network

dlnetwork object

Updated network, returned as an uninitialized dlnetwork object.

To initialize the learnable parameters of a dlnetwork object, use the initialize function.

The replaceLayer function does not preserve quantization information. If the input network is a quantized network, then the output network does not contain quantization information.

Version History

Introduced in R2018b

R2024a: Replace nested layers

Starting in R2024a, you can now replace a layer nested inside a networkLayer or inside the property of a custom layer by specifying the path to the nested layer. To replace a nested layer, use the new layerPath input.

R2024a: LayerGraph objects are not recommended

Starting in R2024a, LayerGraph objects are not recommended. Use dlnetwork objects instead. This recommendation means that these syntaxes are not recommended for LayerGraph input:

• lgraphUpdated = replaceLayer(lgraph,layerName,layers)

• lgraphUpdated = replaceLayer(lgraph,layerName,layers,ReconnectBy=mode)

Most functions that support LayerGraph objects also support dlnetwork objects. This table shows some typical usages of LayerGraph objects and how to update your code to use dlnetwork object functions instead.

Not Recommended	Recommended
lgraph = layerGraph;	net = dlnetwork;
lgraph = layerGraph(layers);	net = dlnetwork(layers,Initialize=false);
lgraph = layerGraph(net);	net = dag2dlnetwork(net);
lgraph = addLayers(lgraph,layers);	net = addLayers(net,layers);
lgraph = removeLayers(lgraph,layerNames);	net = removeLayers(net,layerNames);
lgraph = replaceLayer(lgraph,layerName,layers);	net = replaceLayer(net,layerName,layers);
lgraph = connectLayers(lgraph,s,d);	net = connectLayers(net,s,d);
lgraph = disconnectLayers(lgraph,s,d);	net = disconnectLayers(net,s,d);
plot(lgraph);	plot(net);

To train a neural network specified as a dlnetwork object, use the trainnet function.