Use Experiment Manager Templates for Signal Processing Workflows

Signal Processing Toolbox™ provides preconfigured templates for Experiment Manager (Deep Learning Toolbox) that you can use to design and run deep learning experiments for signal processing workflows.

To learn how to use the templates, follow these examples that show how to conduct experiments on electrocardiogram (ECG) signals:

• Signal Segmentation by Sweeping Hyperparameters — Set up and train a network to divide ECG signals into P-wave, QRS-complex, and T-wave beat morphologies.

• Signal Classification by Sweeping Hyperparameters — Set up and train a classifier to determine if ECG signals exhibit arrhythmia, congestive heart failure, or normal sinus rhythm.

• Signal Classification Using Transfer Learning — Use a pretrained classifier to determine if ECG signals exhibit arrhythmia, congestive heart failure, or normal sinus rhythm.

• Signal Regression by Sweeping Hyperparameters — Set up and train a model to denoise ECG test signals with similar characteristics to a set of signals provided for training.

Experiment Manager enables you to train networks under different conditions and compare results as you search for the specifications that best solve your problem. By sweeping a range of hyperparameter values or by using Bayesian optimization, you can select the most effective signal features to extract and find the optimal architecture and parameters for your network. Use the built-in trainnet (Deep Learning Toolbox) function or define your own custom training function. Monitor your progress by using training plots. Use confusion matrices and custom metric functions to evaluate your trained network. Export trained networks, training information, and experiment results to the MATLAB^® workspace.

You must have a Deep Learning Toolbox™ license to use the Experiment Manager templates for these workflows.