Question about dense re-feeding. about deeprecommender HOT 3 CLOSED

"For model input X and label y both will be equal to B"
-- to be more precise, B=[X,y]
"Let's say model gives y_hat as output for our batch B"
-- the model (f) consumes only X from B, so y_hat=f(x) and we compute loss between y_hat and y
"Now my question is what will be the labels in dense re-feeding? Will it be y_hat or is it B?"
-- In dense re-feeding we feed y_hat again and get y_hat2=f(y_hat), the labels remain the same - y and we compute loss between y_hat2 and y.

from deeprecommender.

Thanks @okuchaiev

from deeprecommender.

@okuchaiev Why do we do Iterative Dense refeeding?

from deeprecommender.