MATLAB Answers

what is recommended way to apply incremental ANN learning

3 views (last 30 days)
Michal Kvasnicka
Michal Kvasnicka on 3 Feb 2020
I would like to apply ANN for approximation (regression) of the larga data set of points {x_i, y_i}, where x_i are input vectors (dimension 30-60), y_i are output scalars - targets and i =1,..., 1e8 or 1e9.
There is an "incremental" learning approach, which is applied on separate parts of data (batches with proper size of data) and partially trained NN weights are used for another part of data (batch) like that:
for bn = 1:num_batches
inputs = <get batch bn inputs>;
targets = <get batch bn targets>;
net = train(net, inputs, targets);
end
Is there any (more detailed) description of the "incremental learning" methodology?
What training algorithm is suitable for incremental training regime?
What ANN structure (number of hidden layers and its size) is recomended?
Any examples or tutorials?
I will be very happy for any relevant recommendation...

Answers (0)


Translated by