可视化深度神经网络

绘制训练进度、评估准确度、解释预测以及将图像网络学习的特征可视化

使用内置的网络准确度和损失图监控训练进度。使用可视化方法，如 Grad-CAM、遮挡敏感度、LIME 和 Deep Dream，研究经过训练的网络。

App

设计和可视化深度学习网络

函数

绘图

`analyzeNetwork`	Analyze deep learning network architecture
`trainingProgressMonitor`	Monitor and plot training progress for deep learning custom training loops (自 R2022b 起)
`updateInfo`	Update information values for custom training loops (自 R2022b 起)
`recordMetrics`	Record metric values for custom training loops (自 R2022b 起)
`groupSubPlot`	Group metrics in training plot (自 R2022b 起)
`plot`	绘制神经网络架构

特征可视化

`testnet`	Test deep learning neural network (自 R2024b 起)
`predict`	Compute deep learning network output for inference
`minibatchpredict`	Mini-batched neural network prediction (自 R2024a 起)
`scores2label`	Convert prediction scores to labels (自 R2024a 起)
`deepDreamImage`	Visualize network features using deep dream
`occlusionSensitivity`	Explain network predictions by occluding the inputs
`imageLIME`	Explain network predictions using LIME (自 R2020b 起)
`gradCAM`	Explain network predictions using Grad-CAM (自 R2021a 起)
`confusionchart`	Create confusion matrix chart for classification problem
`sortClasses`	Sort classes of confusion matrix chart
`rocmetrics`	Receiver operating characteristic (ROC) curve and performance metrics for binary and multiclass classifiers (自 R2022b 起)
`addMetrics`	Compute additional classification performance metrics (自 R2022b 起)
`average`	Compute performance metrics for average receiver operating characteristic (ROC) curve in multiclass problem (自 R2022b 起)

属性

ConfusionMatrixChart Properties	Confusion matrix chart appearance and behavior
ROCCurve Properties	Receiver operating characteristic (ROC) curve appearance and behavior (自 R2022b 起)

主题

可解释性

Deep Learning Visualization Methods
Learn about and compare deep learning visualization methods.
Understand Network Predictions Using Occlusion
This example shows how to use occlusion sensitivity maps to understand why a deep neural network makes a classification decision.
Investigate Spectrogram Classifications Using LIME
This example shows how to use locally interpretable model-agnostic explanations (LIME) to investigate the robustness of a deep convolutional neural network trained to classify spectrograms.
Investigate Classification Decisions Using Gradient Attribution Techniques
This example shows how to use gradient attribution maps to investigate which parts of an image are most important for classification decisions made by a deep neural network.
使用类激活映射调查网络预测
此示例说明如何使用类激活映射 (CAM) 来调查和解释用于图像分类的深度卷积神经网络的预测。
Visualize Image Classifications Using Maximal and Minimal Activating Images
This example shows how to use a data set to find out what activates the channels of a deep neural network.
View Network Behavior Using tsne
This example shows how to use the tsne function to view activations in a trained network.
可视化卷积神经网络的激活
此示例说明如何将图像馈送到卷积神经网络并显示网络的不同层的激活。通过将激活区域与原始图像进行比较，检查激活并发现网络学习的特征。发现较浅层中的通道学习颜色和边缘等简单特征，而较深层中的通道学习眼睛等复杂特征。以这种方式识别特征可以帮助您了解网络学习的内容。
可视化卷积神经网络的特征
此示例说明如何可视化卷积神经网络学习的特征。

训练进度和性能

使用深度学习对网络摄像头图像进行分类
此示例说明如何使用预训练的深度卷积神经网络 GoogLeNet 实时对来自网络摄像头的图像进行分类。
监控深度学习训练进度
此示例说明如何监控深度学习网络的训练进度。
Monitor Custom Training Loop Progress
Track and plot custom training loop progress.
Monitor GAN Training Progress and Identify Common Failure Modes
Learn how to diagnose and fix some of the most common failure modes in GAN training.
ROC Curve and Performance Metrics
Use rocmetrics to examine the performance of a classification algorithm on a test data set.
Compare Deep Learning Models Using ROC Curves
This example shows how to use receiver operating characteristic (ROC) curves to compare the performance of deep learning models.

精选示例

Detect Vanishing Gradients in Deep Neural Networks by Plotting Gradient Distributions

Monitor vanishing gradients while training a deep neural network.

打开实时脚本

Explore Network Predictions Using Deep Learning Visualization Techniques

Investigate network predictions using deep learning visualization techniques.

打开实时脚本

Grad-CAM Reveals the Why Behind Deep Learning Decisions

Use the gradient-weighted class activation mapping (Grad-CAM) technique to understand why a deep learning network makes its classification decisions. Grad-CAM, invented by Selvaraju and coauthors [1], uses the gradient of the classification score with respect to the convolutional features determined by the network in order to understand which parts of the image are most important for classification. This example uses the GoogLeNet pretrained network for images.

打开实时脚本

Interpret Deep Learning Time-Series Classifications Using Grad-CAM

Use the gradient-weighted class activation mapping (Grad-CAM) technique to understand the classification decisions of a 1-D convolutional neural network trained on time-series data.