公开(公告)号：US20230273932A1

公开(公告)日：2023-08-31

申请号：US17947659

申请日：2022-09-19

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

Inventor： Xu LI ,  Yunfeng CAI ,  Mingming SUN ,  Ping LI

IPC: G06F16/2458 ,  G06F16/22

CPC classification number: G06F16/2465 ,  G06F16/2237

Abstract: A method for discovering causality from data includes acquiring to-be-processed data, and obtaining a covariance matrix of the to-be-processed data; determining a first target column in the covariance matrix, taking the number of columns of the first target column as a first place in a rearrangement sequence, and obtaining a first upper triangular matrix according to the first target column; determining a position of the number of columns of the covariance matrix other than the first target column except the first place in the rearrangement sequence according to the first target column and the first upper triangular matrix, and obtaining an upper triangular matrix in each position determination; obtaining an adjacency matrix according to an upper triangular matrix and a rearrangement sequence obtained in final position determination; and generating directed acyclic graph (DAG) by using the adjacency matrix, and taking the DAG as causality discovery result of the to-be-processed data.

公开(公告)号：US20250095250A1

公开(公告)日：2025-03-20

申请号：US18749438

申请日：2024-06-20

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

Inventor： Haoran WANG ,  Zeke XIE ,  Yunfeng CAI ,  Mingming SUN

IPC: G06T11/60 ,  G06T9/00 ,  G06V10/44

Abstract: A method is provided that includes: obtaining a reference image and a description text; extracting a text feature of the description text; and performing the following operations based on a pre-trained diffusion model to generate a target image: in each time step of the diffusion model: calculating a first cross-attention feature of a first image feature and the text feature; obtaining a second cross-attention feature of a second image feature of the reference image and the text feature; editing the first cross-attention feature based on the second cross-attention feature to obtain a third cross-attention feature; and generating a result image feature of the time step based on the third cross-attention feature and the text feature; and decoding a result image feature of a last time step to generate the target image.