Train using MATLAB implement on Arduino

Can I train my model using MATLAB then just implement my model on an Arduino without Tensorflowlite? I imagine this as arrays of weights in my arduino whose number and size depends on the layers and neurons.

If it can be done is it just not as computationally efficient using tensorflowlite?

Yes, this can be done. And it can be just as computationally efficient as TFLite (or even more since you don’t have the overhead of having to manage a graph).

I made an extremely simple demo of this (not trained with MATLAB but the idea is the same):