4
$\begingroup$

I understand how LSTM can be useful for time-series or a series of words in order, but I'm curious if LSTM would provide benefit when the ordering of the inputs is not controlled.

For example, assume I'm trying to make a binary classification decision. My input is a list of 10 User IDs fed through an embedding, and that embedding sequence is fed into an LSTM and out into a single dense output node. If those user IDs can appear in any order, does using RNN/LSTM provide any value?

$\endgroup$
4
$\begingroup$

Oriol Vinyals has played around with this question. He wrote a paper about it here. https://arxiv.org/abs/1511.06391

The short answer is that yes you can use LSTMs for sets that have no order but that you should still pay attention how you order the inputs.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.