Network overfitting processing

[roberty...@gmail.com]

Hello,

I am using the finetune training to train my model for the chi_sim language with the network of [1,48,0,1 Ct3,3,16 Mp3,3 Lfys64 Lfx96 Lrx96 Lfx512 O1c1]


After analyzing this network, I cannot find the any regularization operations in the layers, and there is only one convolution layer in the network.

Then how can I optimize the network structure, such as adding the regularization operations, to avoid the overfiting for the data training? Or any other operations such as extending the network depth?



Thanks for your helpness.

[roberty...@gmail.com]

On the other side, the network contains the LSTM layers.

Does the LSTM in the network train the word order? But I find that the word order in the trained_text file is chaotic.




在 2017年9月18日星期一 UTC+8下午2:30:33，roberty...@gmail.com写道：