apr 21, 2016 - Accelerating convolutional neural networks for mobile applications. (ACM 2016)

Description:

Three-component decomposition

继续拆分 Tucker 分解留下的第二个张量（即 w*h 矩阵）, 拆分后重构可形成块对角张量(block diagonal tensor)

proposed to approximate the original weight
tensor by the sum of some smaller subtensors, each of which is in the Tucker decomposition format.

By rearranging these subtensors, the BTD can be seen as a Tucker decomposition where the second decomposed tensor is a block diagonal tensor.

proposed a Block-Term Decomposition (BTD) method based on low-rank and group sparse decomposition.

结果：
achieved a 7.4% actual speedup for the VGG-16 model, at a cost of a 1.3% increased in the top-5 error.

Added to timeline:

Date: