CNN inference acceleration using dictionary of centroids
This addresses efficiency issues for CNN-chip designers and similar applications where calculation flow can be modified, though it is incremental as it builds on quantization approaches.
The paper tackles the problem of high computational cost and storage requirements for CNN inference by proposing a centroid filter quantization method, achieving 2.9 times better computational gain compared to CP tensor decomposition on ImageNet.
It is well known that multiplication operations in convolutional layers of common CNNs consume a lot of time during inference stage. In this article we present a flexible method to decrease both computational complexity of convolutional layers in inference as well as amount of space to store them. The method is based on centroid filter quantization and outperforms approaches based on tensor decomposition by a large margin. We performed comparative analysis of the proposed method and series of CP tensor decomposition on ImageNet benchmark and found that our method provide almost 2.9 times better computational gain. Despite the simplicity of our method it cannot be applied directly in inference stage in modern frameworks, but could be useful for cases calculation flow could be changed, e.g. for CNN-chip designers.