US 12,229,667 B2
Method and apparatus for generating shared encoder
Daxiang Dong, Beijing (CN); Wenhui Zhang, Beijing (CN); Zhihua Wu, Beijing (CN); Dianhai Yu, Beijing (CN); Yanjun Ma, Beijing (CN); and Haifeng Wang, Beijing (CN)
Assigned to BEIJING BAIDU NETCOM SCIENCE AND TECHNOLOGY CO., LTD, Beijing (CN)
Filed by BEIJING BAIDU NETCOM SCIENCE AND TECHNOLOGY CO., LTD., Beijing (CN)
Filed on Mar. 23, 2021, as Appl. No. 17/209,576.
Application 17/209,576 is a continuation of application No. PCT/CN2020/083551, filed on Apr. 7, 2020.
Claims priority of application No. 201911155509.2 (CN), filed on Nov. 22, 2019.
Prior Publication US 2021/0209417 A1, Jul. 8, 2021
Int. Cl. G06N 3/08 (2023.01); G06F 9/50 (2006.01); G06F 18/214 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 9/5027 (2013.01); G06F 9/5072 (2013.01); G06F 18/2148 (2023.01)] 15 Claims
OG exemplary drawing
 
1. A method for generating a shared encoder, comprising:
sending by a master node a shared encoder training instruction comprising a type of a target shared encoder to be trained to child nodes, to allow each child node obtains training samples based on the type;
sending an initial parameter set of the target shared encoder to be trained to each child node after obtaining a confirmation message returned by each child node, so that the initial parameter set is trained by each child node with the training samples;
obtaining a number of the training samples included in each child node;
determining a weight of each child node based on the number of the training samples included in each child node, wherein the weight is used to balance an importance of each child node in a training of the target shared encoder:
obtaining an updated parameter set of the target shared encoder returned by each child node; and
determining a product of the weight of each child node and a corresponding parameter in the updated parameter set as a value for each parameter in an updated parameter set after modification, and determining a target parameter set corresponding to the target shared encoder by fusing the updated parameter set after modification of each child node based on a first preset rule.