Transfer Learning: How to take certain layers from a pre-trained neural net, add it to a new net, and keep its variables fixed?

Asked Aug 24 '17 at 02:58

Active Nov 14 '17 at 15:22

Viewed 125 times

How can one copy specified layers from a pre-trained neural network and add them to another trained neural net?

For instance,

net1 = NetModel["LeNet Trained on MNIST Data"]

and net2 has a similar structure but it has been trained on a different dataset (however, to keep things simple for this question we can set net2 = net1). How should one go about taking layers 4, 5, 6 from net1 and replace layers 4, 5, 6 in net2, fix their variable so they would not change during the training process, and train net2 on some dataset?

edited Nov 14 '17 at 15:22

J. M.'s missing motivation

124,525
11
401
574

asked Aug 24 '17 at 02:58

Miladiouss

1,883
11
26

1

Does this or this answer your question? – xslittlegrass Aug 24 '17 at 03:05
@xslittlegrass, The first link help Thanks. Should I delete this question then? – Miladiouss Aug 24 '17 at 04:51
How can I transfer layers WITHOUT their variables, i.e. just the layer structure? – Miladiouss Aug 24 '17 at 04:52

Transfer Learning: How to take certain layers from a pre-trained neural net, add it to a new net, and keep its variables fixed?

0 Answers0