New initial starting point (input and output) of already trained LSTM Network

I have input data X and output data Y.

I am training a LSTM network using:

net = trainNetwork(X(1:500), Y(1:500), layers, options);

This trains and initialize the network

However is there a way to initialize the network with for example X(1:600) and Y(1:600), not by retraining but by using the previous trained network ansd start any new predictions from that point on (601 and up)?

Answer 1

[Hornett]

Hi Leon,

Yes it is possible to use an already trained network for new predictions, take a look at the following documentation of transfer learning.

• Get started with transfer learning: https://www.mathworks.com/help/deeplearning/gs/get-started-with-transfer-learning.html

Hope it helps!