Fig. 6From: Neural network training with limited precision and asymmetric exponentWeights exponent values utilization during training of the network (per training batch) [44]Back to article page