How can I continue training with additional data to an already existed neural network?

I have a trained neural network with some data set, let A. That is, the network is trained on data set A. Later on, I want to train this network which is trained on data set A with some additional data set, let B. I think, making together data set A and B, and then training the network is possible. But, what I want is not training for the merged data set, rather continue training on the existed trained network for data set B only. Any help for this... contact me via [e-mail]. Thanks,

Answer 1

[Greg Heath]

In general, this technique will not work if you want to preserve good performance on A.

If you don't want to use all of A, use a subset that exemplifies the basic characteristics of A.

There is a huge history of NN forgetting. Don't waste your time with your original idea.

Spend your time thinking about how to find that characterization subset. Unsupervised clustering of A is one idea. Then use the A cluster centers with B.

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[Roberto de Freitas Cabral]

Although I'm new to neural networks and MATLAB, I had the same question recently.

Searching the web, I've found an interesting link that certainly has to do with your question: https://www.mathworks.com/help/nnet/ug/neural-network-training-concepts.html

I hope it helps.