The Sokal hoax comes to machine learning? With all the screwy machine learning papers out there with unexpected results it's amazing that someone hasn't made a joke paper about some totally silly neutral net tweak that somehow works.