Build Deep Neural Networks

Build networks using command-line functions or interactively using the Deep Network Designer app

Build networks from scratch using MATLAB^® code or interactively using the Deep Network Designer app. Use built-in layers to construct networks for tasks such as classification and regression. To see a list of built-in layers, see List of Deep Learning Layers. You can then analyze your network to understand the network architecture and check for problems before training.

If the built-in layers do not provide the layer that you need for your task, then you can define your own custom deep learning layer. You can specify a custom loss function using a custom output layers and define custom layers with or without learnable parameters. After defining a custom layer, you can check that the layer is valid, GPU compatible, and outputs correctly defined gradients.

For models that cannot be specified as networks of layers, you can define the model as a function. For an example showing how to train a deep learning model defined as a function, see Train Network Using Model Function.

Apps

Deep Network Designer

Design and visualize deep learning networks

Funktionen

alle erweitern

Built-In Layers

Input Layers

`imageInputLayer`	Image input layer
`image3dInputLayer`	3-D image input layer
`sequenceInputLayer`	Sequence input layer
`featureInputLayer`	Feature input layer (Seit R2020b)

Convolution and Fully Connected Layers

`convolution2dLayer`	2-D convolutional layer
`convolution3dLayer`	3-D convolutional layer
`groupedConvolution2dLayer`	2-D grouped convolutional layer
`transposedConv2dLayer`	Transposed 2-D convolution layer
`transposedConv3dLayer`	Transposed 3-D convolution layer
`fullyConnectedLayer`	Fully connected layer

Recurrent Layers

`lstmLayer`	Long short-term memory (LSTM) layer for recurrent neural network (RNN)
`bilstmLayer`	Bidirectional long short-term memory (BiLSTM) layer for recurrent neural network (RNN)
`gruLayer`	Gated recurrent unit (GRU) layer for recurrent neural network (RNN) (Seit R2020a)
`lstmProjectedLayer`	Long short-term memory (LSTM) projected layer for recurrent neural network (RNN) (Seit R2022b)
`gruProjectedLayer`	Gated recurrent unit (GRU) projected layer for recurrent neural network (RNN) (Seit R2023b)

Transformer Layers

`selfAttentionLayer`	Self-attention layer (Seit R2023a)
`attentionLayer`	Dot-product attention layer (Seit R2024a)
`positionEmbeddingLayer`	Position embedding layer (Seit R2023b)
`sinusoidalPositionEncodingLayer`	Sinusoidal position encoding layer (Seit R2023b)
`embeddingConcatenationLayer`	Embedding concatenation layer (Seit R2023b)
`indexing1dLayer`	1-D indexing layer (Seit R2023b)

Neural ODE Layers

neuralODELayer Neural ODE layer (Seit R2023b)

Activation Layers

`reluLayer`	Rectified Linear Unit (ReLU) layer
`leakyReluLayer`	Leaky Rectified Linear Unit (ReLU) layer
`preluLayer`	Parametrized Rectified Linear Unit (PReLU) layer (Seit R2024a)
`clippedReluLayer`	Clipped Rectified Linear Unit (ReLU) layer
`eluLayer`	Exponential linear unit (ELU) layer
`tanhLayer`	Hyperbolic tangent (tanh) layer
`swishLayer`	Swish layer (Seit R2021a)
`geluLayer`	Gaussian error linear unit (GELU) layer (Seit R2022b)
`softmaxLayer`	Softmax layer
`sigmoidLayer`	Sigmoid layer (Seit R2020b)
`functionLayer`	Function layer (Seit R2021b)

Normalization Layers

`batchNormalizationLayer`	Batch normalization layer
`groupNormalizationLayer`	Group normalization layer (Seit R2020b)
`instanceNormalizationLayer`	Instance normalization layer (Seit R2021a)
`layerNormalizationLayer`	Layer normalization layer (Seit R2021a)
`crossChannelNormalizationLayer`	Channel-wise local response normalization layer

Utility Layers

`dropoutLayer`	Dropout layer
`crop2dLayer`	2-D crop layer
`crop3dLayer`	3-D crop layer (Seit R2019b)
`networkLayer`	Network Layer (Seit R2024a)

Data Manipulation

`sequenceFoldingLayer`	(Not recommended) Sequence folding layer
`sequenceUnfoldingLayer`	(Not recommended) Sequence unfolding layer
`flattenLayer`	Flatten layer

Pooling and Unpooling Layers

`averagePooling2dLayer`	Average pooling layer
`averagePooling3dLayer`	3-D average pooling layer
`globalAveragePooling2dLayer`	2-D global average pooling layer (Seit R2019b)
`globalAveragePooling3dLayer`	3-D global average pooling layer (Seit R2019b)
`globalMaxPooling2dLayer`	Global max pooling layer (Seit R2020a)
`globalMaxPooling3dLayer`	3-D global max pooling layer (Seit R2020a)
`maxPooling2dLayer`	Max pooling layer
`maxPooling3dLayer`	3-D max pooling layer
`maxUnpooling2dLayer`	Max unpooling layer

Combination Layers

`additionLayer`	Addition layer
`multiplicationLayer`	Multiplication layer (Seit R2020b)
`concatenationLayer`	Concatenation layer
`depthConcatenationLayer`	Depth concatenation layer

Network Building

`dlnetwork`	Deep learning neural network (Seit R2019b)
`imagePretrainedNetwork`	Pretrained neural network for images (Seit R2024a)
`resnetNetwork`	2-D residual neural network (Seit R2024a)
`resnet3dNetwork`	3-D residual neural network (Seit R2024a)
`addLayers`	Add layers to neural network
`removeLayers`	Remove layers from neural network
`replaceLayer`	Replace layer in neural network
`getLayer`	Look up a layer by name or path (Seit R2024a)
`connectLayers`	Connect layers in neural network
`disconnectLayers`	Disconnect layers in neural network
`expandLayers`	Expand network layers (Seit R2024a)
`groupLayers`	Group layers into network layers (Seit R2024a)
`analyzeNetwork`	Analyze deep learning network architecture
`addInputLayer`	Add input layer to network (Seit R2022b)
`initialize`	Initialize learnable and state parameters of a `dlnetwork` (Seit R2021a)
`networkDataLayout`	Deep learning network data layout for learnable parameter initialization (Seit R2022b)
`setL2Factor`	Set L₂ regularization factor of layer learnable parameter
`getL2Factor`	Get L₂ regularization factor of layer learnable parameter
`setLearnRateFactor`	Set learn rate factor of layer learnable parameter
`getLearnRateFactor`	Get learn rate factor of layer learnable parameter
`dag2dlnetwork`	Convert `SeriesNetwork` and `DAGNetwork` to `dlnetwork` (Seit R2024a)

Network Information

`plot`	Plot neural network architecture
`summary`	Print network summary (Seit R2022b)
`analyzeNetwork`	Analyze deep learning network architecture
`checkLayer`	Check validity of custom or function layer
`isequal`	Check equality of neural networks (Seit R2021a)
`isequaln`	Check equality of neural networks ignoring `NaN` values (Seit R2021a)

Themen

Built-In Layers

Long Short-Term Memory Neural Networks
Learn about long short-term memory (LSTM) neural networks.
Create Simple Deep Learning Neural Network for Classification
This example shows how to create and train a simple convolutional neural network for deep learning classification.
List of Deep Learning Layers
Discover all the deep learning layers in MATLAB.
Build Networks with Deep Network Designer
Interactively build and edit deep learning networks in Deep Network Designer.
Example Deep Learning Networks Architectures
This example shows how to define simple deep learning neural networks for classification and regression tasks.
Generate MATLAB Code from Deep Network Designer
Generate MATLAB code to recreate designing a network in Deep Network Designer.

Custom Layers

Define Custom Deep Learning Layers
Learn how to define custom deep learning layers.
Deep Learning Network Composition
Define custom layers that contain neural networks.
Define Nested Deep Learning Layer Using Network Composition
This example shows how to define a nested custom deep learning layer.
Check Custom Layer Validity
Learn how to check the validity of custom deep learning layers.
View Autogenerated Custom Layers Using Deep Network Designer
This example shows how to import a pretrained TensorFlow™ network and view the autogenerated layers in Deep Network Designer.

Enthaltene Beispiele

Sequence Classification Using Deep Learning

Classify sequence data using a long short-term memory (LSTM) network.

Live Script öffnen

Convert Classification Network into Regression Network

Convert a trained classification network into a regression network.

Live Script öffnen

Sequence Classification Using 1-D Convolutions

Classify sequence data using a 1-D convolutional neural network.

Live Script öffnen

Build Image-to-Image Regression Network Using Deep Network Designer

Use Deep Network Designer to construct an image-to-image regression network for super resolution.

Live Script öffnen

Multilabel Image Classification Using Deep Learning

Use transfer learning to train a deep learning model for multilabel image classification.

Live Script öffnen

Train Neural ODE Network

Train an augmented neural ordinary differential equation (ODE) network.

Live Script öffnen

Erste Schritte mit Deep Network Designer

Das folgende Beispiel veranschaulicht, wie Sie mithilfe von Deep Network Designer ein vortrainiertes GoogLeNet-Netz für die Klassifizierung einer neuen Bildersammlung anpassen können. Dieser Prozess wird „Transfer Learning“ genannt und ist in der Regel wesentlich schneller und einfacher als das Trainieren eines neuen Netzes, da Sie erlernte Merkmale mithilfe einer geringeren Anzahl an Trainingsbildern auf eine neue Aufgabe anwenden können. Bereiten Sie ein Netz mithilfe von Deep Network Designer interaktiv auf Transfer Learning vor.

Live Script öffnen

Erstellen eines einfachen Netzes zur Bildklassifizierung mithilfe von Deep Network Designer

Dieses Beispiel veranschaulicht, wie Sie ein einfaches Convolutional Neural Network für die Deep-Learning-Klassifizierung mithilfe von Deep Network Designer erstellen und trainieren können. Convolutional Neural Networks sind wichtige Werkzeuge für das Deep Learning und eignen sich vor allem für die Bilderkennung.

Live Script öffnen