AM5749: TIDL model training on the fly

Hi Tony,

Typically retraining the model requires retraining with a desktop-class CPU/ GPU (your arrow would need to loop through all three blocks).

Where we run into issues is of course the training time as datasets and network size increases, the retraining time becomes unreasonable at the edge, and would occupy most of the system resources.  Do you have a particular use case that would benefit from retraining at the edge?  

Regards,
Mike Erdahl