Revisiting the Train Loss: an Efficient Performance Estimator for Neural Architecture Search

Binxin Ru\* , Clare Lyle\* , Lisa Schut , Mark van der Wilk , Yarin Gal