bbbdylan / proda

[CVPR2022] PyTorch re-implementation of Prompt Distribution Learning
15 stars 1 forks source link

Why is L_upper in Eq. (9) implemented as a cross-entropy function? #1

Closed tgyy1995 closed 1 year ago

tgyy1995 commented 1 year ago

Thank you for your great work.Not sure if I understand correctly , why is L_upper in Eq. (9) implemented as a cross-entropy function?

lynlynlyn commented 1 year ago

L_upper (Eq. 6) can be considered as adjusting the output logits (by adding a covariance term). Note that the covariance term of the positive term is zero.

tgyy1995 commented 1 year ago

I am still a little confused, can you tell me where is the code to calculate the second term of the denominator in Eq. (6)( (z_i)^TAz_i )? Thank you!

tgyy1995 commented 1 year ago

I understand, thank you!