Can the deployed model be further trained directly on a microcontroller to obtain more classes?

Help
#1

Hello,
Assuming the model has been trained on the platform and deployed to the microcontroller, functioning properly for classification. Now, I am considering adding more classes by collecting new data and training directly on the microcontroller to update the model and obtain more classes. I am currently unsure if this is feasible and would appreciate everyone’s feedback. If this is possible, how should I go about implementing this functionality?
Thanks，
James

#2

No you cannot currently train on device, Edge Impulse models are trained in the cloud before being compiled and deployed. Inference is on device. Additional classes can be added on your project and retrained.

Best

Eoin

1 Like
#3

Hi@Eoin
Thanks for your reply.I understand the situation and thank you very much.

Best

james

#4

Hi@Eoin
The model I trained is based on data from pressure sensors, and each user’s pressure sensor data is different. However, the corresponding models and configurations are the same. If we need to produce many such products, does each product have to be trained and deployed separately on the platform? Is there no optimization method available? I believe this process presents challenges in practical applications, or do you have other methods to simplify this step? In practical applications, the input data for the model is always changing. If each product needs to be trained separately on the platform using the corresponding data, it becomes too complex. Do you have good solutions to address this issue in large-scale production?
Best

James