no code implementations • ICLR 2019 • Haifeng Qian, Mark N. Wegman
This paper proposes a class of well-conditioned neural networks in which a unit amount of change in the inputs causes at most a unit amount of change in the outputs or any of the internal layers.
