How to deploy SVM on ARM Cortex-M processor

Hi everyone.

I have a project in which I have to deploy a SVM (support vector machine) model into an ARM Cortex-M processor. I have already successfully trained my SVM, but I don't know how to deploy it on my edge device (microcontroller). I know that there is a library for neural network (CMSIS NN), but it has little support, as far as I can see. Can anyone help?

Answer 1

[Walter Roberson]

You do code generation on a https://www.mathworks.com/help/stats/classificationsvm.html ClassificationSVM object using https://www.mathworks.com/help/stats/classreg.learning.classif.compactclassificationsvm.predict.html predict().

In your interactive MATLAB session, you save() the classification model you trained. In the code for use on the deployed machine, you load() the model and predict() using it.

[Nikhilesh Karanam]

Dear Walter Roberson,

Which interactive MATLAB session you mean? Could you please share the link of it? Thanks in advance :)

Regards,

Nikhilesh K

[Walter Roberson]

I am referring to the matlab desktop .

Answer 2

[Micael Coutinho]

Thank you. It worked.

[Nikhilesh Karanam]

Hello Micael Coutinho,

I am happy that it worked for you in successfully deploying your SVM model on an ARM cortex-M MCU. I have the same thing to do but could not figure out how. I have achieved the task on MATLAB, got the model trained with a good validation accuracy of 98%, able to make live predictions also. But I want to deploy my algorithm on to the ARM cortex-M MCU. I would be grateful to you if you could please guide me through how you achieved it step by step? Thanks in advance.

Regards,

Nikhilesh K

[Walter Roberson]

You cannot deploy the training portion of your algorithm . You need to save the trained matrix . Then you create a new program that load() the saved matrix and uses it to predict() and it is that new program that you deploy .

[Nikhilesh Karanam]

Thanks. Well, yes. deploying the training portion is not possible. I have used classification learner App, selected Linear SVM for my project, trained the model got a validation accuracy of 98%. I generated a matlab script from the App and used the function for prediction of new data in the generated script which looks like this:

yfit = trainedClassifier.predictFcn(T2)

I get good results on MATLAB and I am stuck here. Please let me know how I can move forward from this point in generating C code if you have any idea. Thanks :)

[Walter Roberson]

I am not sure. You might have to alter that to use

yfit = predict(trainedClassifier, T2);