Hi,
I have a question about the Eqn 4 and eqn 2 in the paper. Both equations are trying to enrich a single feature vector with global context, but they are doing in different ways, i.e. use Q-K-V attention in Eqn. 2 and graph in Eqn. 4. What is the insight?
Hi,
I have a question about the Eqn 4 and eqn 2 in the paper. Both equations are trying to enrich a single feature vector with global context, but they are doing in different ways, i.e. use Q-K-V attention in Eqn. 2 and graph in Eqn. 4. What is the insight?