公开(公告)号：US11804069B2

公开(公告)日：2023-10-31

申请号：US17397497

申请日：2021-08-09

Applicant: BEIJING BAIDU NETCOM SCIENCE AND TECHNOLOGY CO., LTD.

Inventor： Lu Gan ,  Yan Fu ,  Yangjie Zhou ,  Lianghui Chen ,  Shunnan Xu

IPC: G06V40/16 ,  G06F18/232 ,  G06F18/22 ,  G06F18/25 ,  G06V10/762 ,  G06V10/778 ,  G06V10/82 ,  G06V10/44 ,  G06F18/23 ,  G06N3/045 ,  G06N20/00 ,  G06F18/23213 ,  G06F18/2413 ,  G06N3/08

CPC classification number: G06V40/168 ,  G06F18/22 ,  G06F18/23 ,  G06F18/232 ,  G06F18/23213 ,  G06F18/24137 ,  G06F18/25 ,  G06F18/253 ,  G06N3/045 ,  G06N20/00 ,  G06V10/454 ,  G06V10/762 ,  G06V10/763 ,  G06V10/778 ,  G06V10/82 ,  G06N3/08

Abstract: The disclosure provides an image clustering method and an image clustering apparatus. The method includes: obtaining new images, and clustering the new images to obtain a first cluster; determining a historical cluster similar to the first cluster as a second cluster from existing historical clusters; obtaining a distance between the first cluster and the second cluster; and generating a target cluster by fusing the first cluster and the second cluster based on the distance. In the image clustering method, with the image clustering apparatus of the disclosure, secondary clustering processing performed on the existing historical clusters based on newly added images is not required, new and old clusters are directly fused.

公开(公告)号：US20210398026A1

公开(公告)日：2021-12-23

申请号：US17461979

申请日：2021-08-30

Applicant: BEIJING BAIDU NETCOM SCIENCE AND TECHNOLOGY CO., LTD.

Inventor： Lianghui Chen ,  Yan Fu ,  Yangjie Zhou ,  Jun Fang

IPC: G06N20/20 ,  H04L9/30

Abstract: A method includes: sending, by one or more computers, in response to the number of data providers for federated learning being greater than a first threshold, a data field required for the federated learning to a coordinator, the coordinator comprising a computer; receiving, by one or more computers, from the coordinator, information about one or more data providers comprising the required data field, for determining the data providers comprising the required data field as the remaining data providers, wherein the coordinator stores a data field of each data provider; and performing, by one or more computers, federated learning-based modeling with each of the remaining data providers.

公开(公告)号：US20210234687A1

公开(公告)日：2021-07-29

申请号：US17208788

申请日：2021-03-22

Applicant: Beijing Baidu Netcom Science and Technology Co., Ltd.

Inventor： Yangjie Zhou ,  Lianghui Chen ,  Jun Fang ,  Yan Fu

IPC: H04L9/30 ,  G06N20/20 ,  G06N5/00

Abstract: A method includes training, in collaboration with a plurality of collaborators, a plurality of tree models based on data of user samples shared with the plurality of collaborators; performing feature importance evaluation on the trained tree models for assigning weights to feature columns generated by respective ones of the tree models; in response to a determination that a linear model is to be trained in collaboration with a first collaborator of the plurality of collaborators, inputting data of a first user sample shared with the first collaborator into a first tree model of the plurality of tree models and one or more second tree models of the plurality of tree models to obtain a plurality of one-hot encoded feature columns; and screening the obtained feature columns based on the respective weights and training the linear model according to the screened feature columns and the data of the first user sample.

公开(公告)号：US20210224879A1

公开(公告)日：2021-07-22

申请号：US17152562

申请日：2021-01-19

Applicant: Beijing Baidu Netcom Science and Technology Co., Ltd.

Inventor： Lianghui Chen ,  Yan Fu ,  Quanbin Wang ,  Xiaoxuan Yang ,  Liangang Peng

IPC: G06Q30/06 ,  G06N20/00 ,  G06F16/245

Abstract: A method, electronic device and storage medium for item recommendation and for model training, which relates to the field of artificial intelligence, are disclosed. According to some embodiments: an item feature expression database is created using a pre-trained user-clicking-item task model and at least two pieces of feature information of items in the repository of items to be recommended; a feature expression of a user is obtained using the pre-trained user-clicking-item task model and at least two pieces of feature information of the user; identifiers of N items to be recommended are obtained according to the feature expression of the user and the item feature expression database; relevant information of the N items is recommended to the user based on the identifiers of the N items.