Train NN with new data while maintaining previous training

Hey, After training a network with a large set of data, I want to "adjust" the training with new small incoming data - without resting the weights and biases.

Basically I want to continue the training for every new data to enhance the network - but not to run an entire new NN learning with the old +new data

How can I do that?

Answer 1

[Ameer Hamza]

You are trying to do transfer learning. This is not only possible but quite popular techniques nowadays. Refer to this example to get the idea how to do it in MATLAB.