How do I train an existing shallow neural network with new data?

Hi,

I have a trained shallow NN that I want to retrain using new experimental data in order to improve the generalization capability of the network. I dont want to reinitialize and train the whole network from scratch. I want the network to adapt itself to the new data. If I use the train function, does it forget the weights and biases of the previous network.

How do I do this? Please help.

Answer 1

[Matt J]

No, it does not forget them, as seen from this example, which talks about resuming a previous training attempt,

https://www.mathworks.com/help/deeplearning/ref/network.train.html#btzb5il-1

If it forgot the weights and biases, resuming would not be possible this way.

[Mohammad Mainul]

Is it better to include some parts of the original train set with the new set for re-training? Does it have any effect?

[Matt J]

It would be best to include all of the original training set. Otherwise, the weights will forget the data you omit as the algorithm iterates.