1 paper
Wonho Bae, Yi Ren, Mohamad Osama Ahmed +3
Although neural networks are conventionally optimized towards zero training loss, it has been recently learned that targeting a non-zero training loss threshold, referred to as a f…