Towards a Novel 8-Bit Floating-Point Format to Increase Robustness in Convolutional Neural Networks | Publicación