Model structure

[My-Khan]

Hello ,
Can i represent DRNN in the below structure for Named entity task. any suggestion for improvement will be highly appreciated. Also is it based on Elman Architecture which is described in (Kaisheng Yao et al.,2013) ?

[zhongkaifu]

Yes. This is a basic RNN model structure: single recurrent hidden layer and forward model. It can be used for NER task. For more complicated model, you could try bi-directional model, deep recurrent hidden layer, different recurrent hidden type(BPTT, LSTM), CRF output and so on.

[My-Khan]

many thanks for clarification.
also on the home page of RNNShrap the model structure given their for deep bi-directional RNN-CRF network all the three hidden layer are labeled Hidden layer #1, Hidden layer #1 ,Hidden layer #1 . is this the typing mistake? or its real.

also the equation used from computation is this one
h(t)=σ(U.x(t)+W.h(t-1) ), t=1….T and
y(t)=g(Vh(t))

if yes
then please explain g(Vh(t)).

is same equation is used for computation in basic RNN and DEEP RNN?

[zhongkaifu]

Thanks for pointing out. It's a typo. They should be "Hidden layer #1", "Hidden layer #2", "Hidden layer #3". I will fix it.

g(Vh(t)) is the function to calcuate y(t) by using the output of hidden layer and weights between hidden layer and y.

[My-Khan]

Many thanks.
can i use your diagrams in my paper with reference?

[zhongkaifu]

Sure. I'm glad these diagrams is helpful for your paper.

[My-Khan]

so nice of you

[zhongkaifu]

Fixed it. Thanks for your pointing it out.

[My-Khan]

Appreciated , Is in future there is chance of implementation of Skip-Chain CRF in C# from you?

[zhongkaifu]

Maybe, but currently, I have no plan about it.

[My-Khan]

Ok thanks for kind reponse