Skip to main content
[Preprint]. 2023 Aug 16:arXiv:2301.01885v2. Originally published 2023 Jan 5. [Version 2]
D ∈ {X, y}: dataset
f: model
nfolds: folds for K-fold partitioning
 λ: enhancement step size
for k = 1 : nfolds do
   Establish Dtr, Dheld-out
   Train f
   g ← ∇xA where A = L(f, x) or DF(f, x)
   Xheld-out ← Xheld-out − λg
end for