ManifoldResolved: YesPOLL · none

Can a model have a lower loss on the validation set than it shows during training?