A linear layer can learn scaling. That’s what i said. And here is the proof:

Scaling: is y = x/(max-min)-min/(max-min)

which is in the form of y=w.x+b. With w and b learnable