US 11,704,569 B2
Methods and apparatus for enhancing a binary weight neural network using a dependency tree
Yiwen Guo, Beijing (CN); Anbang Yao, Beijing (CN); Hao Zhao, Beijing (CN); Ming Lu, Beijing (CN); and Yurong Chen, Beijing (CN)
Assigned to Intel Corporation, Santa Clara, CA (US)
Appl. No. 16/615,097
Filed by INTEL CORPORATION, Santa Clara, CA (US)
PCT Filed May 23, 2018, PCT No. PCT/US2018/034088
§ 371(c)(1), (2) Date Nov. 19, 2019,
PCT Pub. No. WO2018/217863, PCT Pub. Date Nov. 29, 2018.
Claims priority of provisional application 62/510,075, filed on May 23, 2017.
Prior Publication US 2020/0167654 A1, May 28, 2020
Int. Cl. G06N 3/08 (2023.01); G06N 3/082 (2023.01); G06N 3/042 (2023.01); G06N 3/044 (2023.01); G06N 5/01 (2023.01)
CPC G06N 3/082 (2013.01) [G06N 3/042 (2023.01); G06N 3/044 (2023.01); G06N 5/01 (2023.01)] 22 Claims
OG exemplary drawing
 
1. A method of enhancing a convolutional neural network (CNN) having binary weights comprising:
constructing a tree for obtained binary tensors, the tree having a plurality of nodes beginning with a root node in each layer of the CNN;
calculating a convolution of an input feature map with an input binary tensor at the root node of the tree; and
searching a next node from the root node of the tree and calculating a convolution at the next node using a previous convolution result calculated at the root node of the tree.