- Abstract: We proposed Additive Powers-of-Two (APoT) quantization, an efﬁcient nonuniform quantization scheme that attends to the bell-shaped and long-tailed distribution of weights in neural networks. By constraining all quantization levels as a sum of several Powers-of-Two terms, APoT quantization enjoys overwhelming efﬁciency of computation and a good match with weights’ distribution. A simple reparameterization on clipping function is applied to generate better-deﬁned gradient for updating of optimal clipping threshold. Moreover, weight normalization is presented to reﬁne the input distribution of weights to be more stable and consistent. Experimental results show that our proposed method outperforms state-of-the-art methods, and is even competitive with the full-precision models demonstrating the effectiveness of our proposed APoT quantization. For example, our 3-bit quantized ResNet-34 on ImageNet only drops 0.3% Top-1 and 0.2% Top-5 accuracy without bells and whistles, while the computation of our model is approximately 2× less than uniformly quantized neural networks.
- Keywords: Quantization, Efficient Inference, Neural Networks