about LSTM
YsSue opened this issue · comments
YsSue commented
Hi ronghang,
I am reading your excellent work "Segmentation from Natural Language Expressions" , I am wondering someting about the Dtext = 1000 dimensional hidden state h, which is obtained by LSTM. Seems like that you encode the word vectors with LSTM, but is there some physical definition of the output h? and how it makes sense on segmentation task?
Thank you very much for your time and consideration. I look forward to hearing from you earlier.