5 mayo 2015 año - Accelerating very deep convolutional networks for classification and detection. (TPAMI 2015)(copy)

Descripción:

Three-component Decomposition

w*h*c*n -> w*h*c*d d*n -> w*1 1*h 1*1

conduct a two-component decomposition on the first weight tensor after the SVD decomposition, which turns into a three- component decomposition method.

结果：
By utilizing this three-component decomposition, only a 0.3% increased top-5 error was produced in for a 4× theoretical speedup.