Fetching the paper…

GXNOR-Net: Training deep neural networks with ternary weights and activations without full-precision memory under a unified discretization framework · Around