Is it possible to train Neural Networks in Simulink?

We have a series of recurrent neural networks that we want to test in simulink. The objective is to apply different controls that will be based on the predictions generated by the Artificial Intelligence models. During the control process the models will be fed with data with which it has not been trained. Therefore, the intention is to be able to train these artificial intelligence models in simulink.

Thanks.

Answer 1

[David Willingham]

Hi Jon,

The Stateful Predict and Stateful Classify blocks can be used to import trainedt recurrent neural networks into Simulink.

[Jon Yarza]

Hi Dadiv,

Thank you for your answer.

However, in some applications the model needs to learn from new data. With those blocks this is not possible right? How could you launch a training for a model from simulink?

Thanks in advance,

Jon

[Lucas García]

Hi Jon,

Adding to what David mentions, once your model has been trained using your available training data, you can bring it to Simulink for system-level simulation and code generation. Here you can see a more concrete example on how to predict responses from a trained neural network in Simulink.

Because model training is a time-consuming task, this is something that is done outside of the simulation environment. So, given the data you have for training the neural network and testing it in simulation comes from the same distribution (this is frequently assumed), you can train the network using your available training data, validate the network, and then integrate it with the rest of your system in Simulink. The neural network block used in Simulink will perform against unseen data, inferring the results from the observed training data.

Lucas