I read the tutorial of LSTM from here. However, I have certain doubts that I need to address.

  1. Since we use true labels and do not remove anything from the original data, then how is it possible for the LSTM model's predicted output to match the real labels as it throws data?

  2. And how do we determine the number of output neurons?

According to my understanding, in word-to-word prediction, one cell's outputs are the number of words (exiting in vocabulary).

  • $\begingroup$ What data does LSTM throws away? $\endgroup$
    – hanugm
    Jul 26, 2021 at 9:00
  • $\begingroup$ I think your doubt is related to self supervised learning. Suppose there is a text on which next word prediction LSTM is trained, then label is the next word itself for the window of words. Suppose your LSTM takes 4 words as input and predicts next word as ouput, then the window of words i am a good girl. Then word girl will be the true label for the input I am a good. $\endgroup$
    – hanugm
    Jul 26, 2021 at 9:16
  • $\begingroup$ @hanugm thank you for the comment. No it's not self-supervision but the actual procedure of LSTM how it works. Somehow I got the concept but still have confused. $\endgroup$ Jul 26, 2021 at 11:57
  • $\begingroup$ @hanugm Please see my this Question ai.stackexchange.com/questions/28843/… $\endgroup$ Jul 26, 2021 at 13:29
  • $\begingroup$ Okay............. $\endgroup$
    – hanugm
    Jul 26, 2021 at 14:37


You must log in to answer this question.