A question about the paper (Order matters: seq2seq for sets)

[AgoloYomnaIsmail]

Hello, I am trying to implement this paper https://arxiv.org/pdf/1511.06391.pdf (order matters:seq2seq for sets) and I have a question regarding your reply to this question https://www.reddit.com/r/MachineLearning/comments/85yjzo/d_order_matters_sequence_to_sequence_for_sets/, which is :

how does the cell state input (q*) be of shape 512 (like ur example) and (q) which is the output of the cell be of size (512/2)?

I am having problem regarding dimensions , because lstm takes q* as input which has a size= size(q)+size(r) , while (q) is the output state of lstm which should have the size of q* . I am using the annotations used in the paper. Thanks.