内置层

使用内置层构建深度神经网络

对于大多数任务，您可以使用内置层。如果没有您的任务所需的内置层，则可以定义您自己的自定义层。您可以定义具有可学习参数和状态参数的自定义层。定义自定义层后，您可以检查该层是否有效，是否与 GPU 兼容，以及是否输出正确定义的梯度。要查看支持的层的列表，请参阅深度学习层列表。

App

深度网络设计器

设计和可视化深度学习网络

函数

全部展开

内置层

输入层

`inputLayer`	输入层 (自 R2023b 起)
`imageInputLayer`	Image input layer
`image3dInputLayer`	3-D image input layer
`sequenceInputLayer`	Sequence input layer
`featureInputLayer`	Feature input layer

卷积和全连接层

`convolution1dLayer`	1-D convolutional layer (自 R2021b 起)
`convolution2dLayer`	2-D convolutional layer
`convolution3dLayer`	3-D convolutional layer
`groupedConvolution2dLayer`	2-D grouped convolutional layer
`transposedConv1dLayer`	Transposed 1-D convolution layer (自 R2022a 起)
`transposedConv2dLayer`	Transposed 2-D convolution layer
`transposedConv3dLayer`	Transposed 3-D convolution layer
`spectralConvolution1dLayer`	1-D spectral convolutional layer (自 R2026a 起)
`spectralConvolution2dLayer`	2-D spectral convolutional layer (自 R2026a 起)
`spectralConvolution3dLayer`	3-D spectral convolutional layer (自 R2026a 起)
`fullyConnectedLayer`	Fully connected layer
`complexFullyConnectedLayer`	Complex fully connected layer (自 R2026a 起)

循环层

`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)
`lstmProjectedLayer`	Long short-term memory (LSTM) projected layer for recurrent neural network (RNN) (自 R2022b 起)
`gruProjectedLayer`	Gated recurrent unit (GRU) projected layer for recurrent neural network (RNN) (自 R2023b 起)

变换器层

`embeddingLayer`	Embedding layer (自 R2026a 起)
`selfAttentionLayer`	Self-attention layer (自 R2023a 起)
`attentionLayer`	Dot-product attention layer (自 R2024a 起)
`positionEmbeddingLayer`	Position embedding layer (自 R2023b 起)
`sinusoidalPositionEncodingLayer`	Sinusoidal position encoding layer (自 R2023b 起)
`embeddingConcatenationLayer`	Embedding concatenation layer (自 R2023b 起)
`indexing1dLayer`	1-D indexing layer (自 R2023b 起)

神经 ODE 层

`neuralODELayer`	Neural ODE layer (自 R2023b 起)
`deep.ode.options.ODE1`	Neural ODE solver options for nonstiff differential equations using Euler method (自 R2025a 起)
`deep.ode.options.ODE45`	Neural ODE solver options for nonstiff differential equations (自 R2025a 起)

激活层

`reluLayer`	修正线性单元 (ReLU) 层
`leakyReluLayer`	泄露修正线性单元 (ReLU) 层
`preluLayer`	Parametrized rectified linear unit (PReLU) layer (自 R2024a 起)
`clippedReluLayer`	裁剪修正线性单元 (ReLU) 层
`eluLayer`	指数线性单元 (ELU) 层
`tanhLayer`	双曲正切 (tanh) 层
`swishLayer`	swish 层
`geluLayer`	高斯误差线性单元 (GELU) 层 (自 R2022b 起)
`softmaxLayer`	Softmax 层
`sigmoidLayer`	sigmoid 层
`softplusLayer`	Softplus layer
`complexReluLayer`	复数修正线性单元 (ReLU) 层 (自 R2025a 起)
`zreluLayer`	ZReLU Layer (自 R2026a 起)
`functionLayer`	Function layer (自 R2021b 起)

归一化层

`batchNormalizationLayer`	Batch normalization layer
`groupNormalizationLayer`	Group normalization layer
`instanceNormalizationLayer`	Instance normalization layer
`inverseNormalizationLayer`	Inverse normalization layer (自 R2026a 起)
`layerNormalizationLayer`	Layer normalization layer
`crossChannelNormalizationLayer`	Channel-wise local response normalization layer

实用工具层

`dropoutLayer`	丢弃层
`spatialDropoutLayer`	Spatial dropout layer (自 R2024a 起)
`flattenLayer`	展平层
`crop2dLayer`	2-D crop layer
`crop3dLayer`	3-D crop layer
`scalingLayer`	Scaling layer
`quadraticLayer`	Quadratic layer
`identityLayer`	Identity layer (自 R2024b 起)
`complexToRealLayer`	Complex-to-real layer (自 R2024b 起)
`realToComplexLayer`	Real-to-complex layer (自 R2024b 起)
`networkLayer`	Network Layer (自 R2024a 起)
`reshapeLayer`	Reshape layer (自 R2025a 起)
`permuteLayer`	Permute layer (自 R2025a 起)

池化和去池化层

`averagePooling1dLayer`	1-D average pooling layer (自 R2021b 起)
`averagePooling2dLayer`	Average pooling layer
`averagePooling3dLayer`	3-D average pooling layer
`adaptiveAveragePooling2dLayer`	Adaptive average pooling 2-D layer (自 R2024a 起)
`globalAveragePooling1dLayer`	1-D global average pooling layer (自 R2021b 起)
`globalAveragePooling2dLayer`	2-D global average pooling layer
`globalAveragePooling3dLayer`	3-D global average pooling layer
`globalMaxPooling1dLayer`	1-D global max pooling layer (自 R2021b 起)
`globalMaxPooling2dLayer`	Global max pooling layer
`globalMaxPooling3dLayer`	3-D global max pooling layer
`maxPooling1dLayer`	1-D max pooling layer (自 R2021b 起)
`maxPooling2dLayer`	Max pooling layer
`maxPooling3dLayer`	3-D max pooling layer
`maxUnpooling2dLayer`	Max unpooling layer

组合层

`additionLayer`	加法层
`multiplicationLayer`	相乘层
`concatenationLayer`	串联层
`depthConcatenationLayer`	深度串联层

网络创建

`dlnetwork`	深度学习神经网络
`imagePretrainedNetwork`	适用于图像的预训练神经网络 (自 R2024a 起)
`resnetNetwork`	2-D residual neural network (自 R2024a 起)
`resnet3dNetwork`	3-D residual neural network (自 R2024a 起)
`dag2dlnetwork`	Convert `SeriesNetwork` and `DAGNetwork` to `dlnetwork` (自 R2024a 起)

网络编辑

`addLayers`	向神经网络添加层
`removeLayers`	从神经网络中删除层
`replaceLayer`	Replace layer in neural network
`getLayer`	Look up a layer by name or path (自 R2024a 起)
`connectLayers`	在神经网络中连接各层
`disconnectLayers`	在神经网络中断开层连接
`expandLayers`	Expand network layers (自 R2024a 起)
`groupLayers`	Group layers into network layers (自 R2024a 起)
`addInputLayer`	Add input layer to network (自 R2022b 起)

可学习参数

`initialize`	初始化神经网络的可学习参数和状态参数
`networkDataLayout`	Deep learning network data layout for learnable parameter initialization (自 R2022b 起)
`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

网络信息

`plot`	绘制神经网络架构
`summary`	打印网络摘要 (自 R2022b 起)
`analyzeNetwork`	Analyze deep learning network architecture
`checkLayer`	Check validity of custom or function layer
`isequal`	Check equality of neural networks
`isequaln`	Check equality of neural networks ignoring `NaN` values

主题

长短期记忆神经网络
了解长短期记忆 (LSTM) 神经网络。
创建简单的深度学习神经网络以用于分类
此示例说明如何创建和训练简单的卷积神经网络来进行深度学习分类。
针对回归训练卷积神经网络
此示例说明如何训练卷积神经网络来预测手写数字的旋转角度。
深度学习层列表
探索 MATLAB^® 中的所有深度学习层。
使用深度网络设计器构建网络
在深度网络设计器中以交互方式构建和编辑深度学习网络。
Create and Train Network with Nested Layers
This example shows how to create and train a network with nested layers using network layers. (自 R2024a 起)
Neural Network Weight Tying
This example shows how to create and edit neural networks that have layers that share learnable parameters. (自 R2026a 起)
Train Neural Network with Weight Tying
This example shows how to train a neural network with weight tying by passing learnable parameters between layers. (自 R2026a 起)
Example Deep Learning Network Architectures
This example shows how to define simple deep learning neural networks for classification and regression tasks.
Choose an AI Model
Explore options for choosing an AI model.
从深度网络设计器生成 MATLAB 代码
生成 MATLAB 代码，以便在深度网络设计器中重新创建设计网络。
多输入和多输出网络
了解如何定义和训练具有多个输入或多个输出的深度学习网络。

精选示例

使用深度学习进行序列分类

此示例说明如何使用长短期记忆 (LSTM) 网络对序列数据进行分类。

打开实时脚本

Build Custom Network for Time Series Modeling of a Virtual Sensor

Use the Time Series Modeler app to build and train a custom network for time series modeling of the nonlinear torque dynamics of a spark-ignition (SI) engine. In this example, you build a custom 1-D convolutional neural network with two branches.