keras-self-attention icon indicating copy to clipboard operation
keras-self-attention copied to clipboard

Published 20 hours ago • verified publisher icon CyberZHG

Question about the SeqSelfAttention.

Open katekats opened this issue 2 years ago • 0 comments

My question is: For the additive self-attention approach, are word embeddings from other timestamps taken into consideration for calculating the attention weights or only from the current timestamp (meaning word embeddings of the current sentence/input)?

katekats avatar May 16 '22 08:05 katekats