Avoid training certain neurons

Question

Hamid Moazed 2019-12-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/497749-avoid-training-certain-neurons

评论： Hamid Moazed 2019-12-23

Using the Deep Learning Toolbox, I wish to construct a simple feed-forward network for a simulation, however assume I have already trained one of the hidden neurons (out of several) with the correct weights and biases and I don't want them to change during training. How can I make this single specific neuron be "constant" and not get retrained with new wights and biases while the rest of the network is being trained?

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

Hiro Yoshino 2019-12-23

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/497749-avoid-training-certain-neurons#answer_407422

在 MATLAB Online 中打开

There is an option to keep specific layers' learning rates low so you can fix them as they are.

for example

fullyConnectedLayer(<outputsize>, 'WeightLearnRateFactor', 0, 'BiasLearnRateFactor', 0)

This way, you would multiply zero to the global learning rate, which is set via trainingOptions function, and thus the learning rates of the weights in the fully-connected-layer are set as zero.

Check this out - this may help you out too.