Comment by Dilettante_

>you could perhaps make an argument that, just like weight decay, an apparent "anti-contribution" moves the learning trajectory along

Was that not the joke?