How do I add layer to a custom NN while keeping the weights and biases of trained layer constant?

Question

Mohammadmahdi Malakizandi 2018-8-27

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/416383-how-do-i-add-layer-to-a-custom-nn-while-keeping-the-weights-and-biases-of-trained-layer-constant

评论： Mohammadmahdi Malakizandi 2018-8-28

Hi, I have trained a custom Neural Network with 3 inputs and 1 output to emulate my system.You can find the code for the emulator at the end. Now I want to add two layers to the network that get two of the inputs(F and Y) and give back the other input(U), in a way that the output of my Neural Network emulator is my desired output. But i want my emulator stay as it is and just the added layers are trained. Thank you in Advance

p1=tonndata(F,true,false); p2=tonndata(Y,true,false); p3=tonndata(U,true,false); t=tonndata(Y_dot,true,false);

net=network; net.numInputs = 3;

net.numLayers = 4;

net.biasConnect(1)=1; net.biasConnect(2)=1; net.biasConnect(3)=1; net.biasConnect(4)=1;

net.inputConnect(1,1)=1; net.inputConnect(2,2)=1; net.inputConnect(3,3)=1;

net.layerConnect(4,1)=1; net.layerConnect(4,2)=1; net.layerConnect(4,3)=1;

net.outputConnect(4)=1;

net.layers{1}.size=10; net.layers{1}.transferFcn='tansig'; net.layers{1}.initFcn='initnw';

net.layers{2}.size=5; net.layers{2}.transferFcn='tansig'; net.layers{2}.initFcn='initnw';

net.layers{3}.size=5; net.layers{3}.transferFcn='tansig'; net.layers{3}.initFcn='initnw';

net.layers{4}.size=3; net.layers{4}.transferFcn='purelin'; net.layers{4}.initFcn='initnw';

net.inputWeights{1,1}.delays=1:12; net.inputWeights{2,2}.delays=1:12; net.inputWeights{3,3}.delays=1:12;

net.initFcn= 'initlay';

net.performFcn='mse';

net.trainFcn='trainbr';

net.trainParam.show=1;

net.trainParam.goal=1e-8; net.trainParam.epochs=200; net.trainParam.show=1;

[net,tr]=train(net,[p1;p2;p3],t);

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Greg Heath 2018-8-27

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/416383-how-do-i-add-layer-to-a-custom-nn-while-keeping-the-weights-and-biases-of-trained-layer-constant#answer_334269

What I have done is to store the signals that will be inputs to the addition.

Then I use that stored data set to train the addition.

Hope this helps.

Thank you for formally accepting my answer

Greg

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Mohammadmahdi Malakizandi 2018-8-28

Thanks for the reply but I think I may have misrepresented what I wanted to do. The problem is I don't know what the inputs should be. So I don't have the signals. What I have is the output of the emulator. I want the output of the new NN be in the way that when I use it as the input of the emulator, the output of the emulator be what I want.

请先登录，再进行评论。

How do I add layer to a custom NN while keeping the weights and biases of trained layer constant?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

回答（1 个）

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

另请参阅

类别

标签

产品

Community Treasure Hunt

How do I add layer to a custom NN while keeping the weights and biases of trained layer constant?

0 个评论 显示 -2更早的评论隐藏 -2更早的评论

回答（1 个）

1 个评论 显示 -1更早的评论隐藏 -1更早的评论

另请参阅

类别

标签

产品

Community Treasure Hunt

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

1 个评论
显示 -1更早的评论隐藏 -1更早的评论