EnergyAttention/Energy-Based-CrossAttention

Query about the conditional energy

zhuqiangLu opened this issue · 0 comments

Hi there, thank you for sharing your wonderful work.

I am a bit confused with Eq(8) conditional energy function w.r.t K in your paper.

Correct me if I am wrong, I understand the energy function Eq(3) from the Morden Hopfield Network, however, by comparing Eq(8) and Eq(3), I noticed that you replaced the query with key. May I ask why this modification enable the energy function conditional to keys?