Fig. 7From: Neural network training with limited precision and asymmetric exponentWeights gradient exponent values utilization during training of the network (per training batch) [44]Back to article page