Title:
NODE FUSION METHOD FOR COMPUTATIONAL GRAPH, AND DEVICE
Document Type and Number:
WIPO Patent Application WO/2022/143419
Kind Code:
A1
Abstract:
Disclosed are a node fusion method for a computational graph, and a device, which can be applied to the field of artificial intelligence, and in particular to a deep learning framework. The method comprises: converting a neural network into a computational graph; extracting one or more parallelizable branch groups from the computational graph on the basis of a dependence relationship between nodes in the computational graph, wherein the dependence relationship is used for indicating at least one of the following relationships: the parallelizable branch groups having a common father node, the parallelizable branch groups having a common child node, the parallelizable branch groups having no father node, and the parallelizable branch groups having no child node; and finally, fusing a plurality of nodes, respectively from different sub-branches, in any parallelizable branch group, so as to obtain a new computational graph. In the present application, the possibility of some other parallelizable branches is taken into consideration, such that parallelizable branch combinations that are different from those limited by rules of the prior art are found, and thus, nodes in these branch combinations can be fused during the node fusion of a computational graph, thereby expanding the range of fusion nodes that can be acquired.
More Like This:
Inventors:
ZHANG ZHAOCHUANG (CN)
GAO XIONG (CN)
ZENG ZITAO (CN)
GAO XIONG (CN)
ZENG ZITAO (CN)
Application Number:
PCT/CN2021/140906
Publication Date:
July 07, 2022
Filing Date:
December 23, 2021
Export Citation:
Assignee:
HUAWEI TECH CO LTD (CN)
International Classes:
G06N3/08
Foreign References:
| CN111260019A | 2020-06-09 | |||
| CN111935005A | 2020-11-13 | |||
| CN109740751A | 2019-05-10 | |||
| CN109543825A | 2019-03-29 | |||
| US20200249998A1 | 2020-08-06 | |||
| CN202011595030A | 2020-12-28 |
Attorney, Agent or Firm:
SHENPAT INTELLECTUAL PROPERTY AGENCY (CN)
Download PDF: