Does the Matlab trainNetwork() function re-initialise the model's weights to zero? Or will it use the previously learnt weights as a starting point?

Really would help to know this so that I know whether or not I am just wasting my time re training my model on the new data I received. Doesn't mention it anywhere in the documentation, and I have found threads on here discussing the train() function in regards to this question, but none on the trainNetwork() function, so I don't know if the answers will be the same.

Answer 1

[Sailesh Kalyanapu]

It is my understading that you are looking to retrain an existing trained model on new data using trainNetwork() function.

It is possible to save networks as .mat files during training and resume training from the last saved checkpoint network.

Please refer to the below documentation and example to Save Checkpoint Networks and Resume Training

• Train deep learning neural network - MATLAB trainNetwork (mathworks.com)
• Resume Training from Checkpoint Network - MATLAB & Simulink (mathworks.com)