论文标题
有效的基于残留编号系统的Winograd卷积
Efficient Residue Number System Based Winograd Convolution
论文作者
论文摘要
先前的研究表明，Winograd算法可以降低卷积神经网络（CNN）的计算复杂性，其权重和激活在浮点上表示。但是，很难将该方案应用于低精度量化（例如INT8）网络的推断。我们的工作将Winograd算法扩展到残留编号系统（RNS）。使用Winograd变换和低成本（例如8位）算术算术，在不降低网络预测准确性的情况下，使用Winograd变换和低成本（例如8位）算术来精确地计算出大型转换图（例如10 x 10至16 x 16）和激活斑块的最小复杂度卷积。对于3 x 3和5 x 5过滤器，算术复杂性的降低最高为7.03倍，而性能改善分别为2.30倍至4.69倍。
Prior research has shown that Winograd algorithm can reduce the computational complexity of convolutional neural networks (CNN) with weights and activations represented in floating point. However it is difficult to apply the scheme to the inference of low-precision quantized (e.g. INT8) networks. Our work extends the Winograd algorithm to Residue Number System (RNS). The minimal complexity convolution is computed precisely over large transformation tile (e.g. 10 x 10 to 16 x 16) of filters and activation patches using the Winograd transformation and low cost (e.g. 8-bit) arithmetic without degrading the prediction accuracy of the networks during inference. The arithmetic complexity reduction is up to 7.03x while the performance improvement is up to 2.30x to 4.69x for 3 x 3 and 5 x 5 filters respectively.
