知搜-全球专利检索

  1. Login
  2. Sign Up

Search Results

19 records (took 0.016 seconds)

    USING MULTIPLE TRAINED MODELS TO REDUCE DATA LABELING EFFORTS
    11.
    发明公开
    USING MULTIPLE TRAINED MODELS TO REDUCE DATA LABELING EFFORTS 审中-公开

    公开(公告)号：US20230350880A1

    公开(公告)日：2023-11-02

    申请号：US18219333

    申请日：2023-07-07

    Applicant: Xerox Corporation

    Inventor： Matthew Shreve ,  Francisco E. Torres ,  Raja Bala ,  Robert R. Price ,  Pei Li

    IPC: G06F16/23 ,  G06N20/00

    CPC classification number: G06F16/2379 ,  G06N20/00

    Abstract: A method of labeling a dataset includes inputting a testing set comprising a plurality of input data samples into a plurality of pre-trained machine learning models to generate a set of embeddings output by the plurality of pre-trained machine learning models. The method further includes performing an iterative cluster labeling algorithm that includes generating a plurality of clusterings from the set of embeddings, analyzing the plurality of clusterings to identify a target embedding with a highest duster quality, analyzing the target embedding to determine a compactness for each of the plurality of clusterings of the target embedding, and identifying a target cluster among the plurality of clusterings of the target embedding based on the compactness. The method further includes assigning pseudo-labels to the subset of the plurality of input data samples that are members of the target duster.

    Employee task verification to video system
    13.
    发明授权
    Employee task verification to video system 有权

    公开(公告)号：US10984355B2

    公开(公告)日：2021-04-20

    申请号：US14689320

    申请日：2015-04-17

    Applicant: Xerox Corporation

    Inventor： Michael C. Mongeon ,  Robert Loce ,  Matthew Shreve

    IPC: G06Q10/06 ,  G06T7/246 ,  H04N21/44 ,  G06F3/01 ,  G06K9/00 ,  H04N5/77

    Abstract: When monitoring a workspace to determine whether scheduled tasks or chores are completed according to a predetermined schedule, a video monitoring system monitors a region of interest (ROI) to identify employee-generated signals representing completion of a scheduled task. An employee makes a mark or gesture in the ROI monitored by the video monitoring system and the system analyzes pixels in each captured frame of the ROI to identify an employee signal, map the signal to a corresponding scheduled task, update the task as having been completed upon receipt of the employee signal, and alert a manager of the facility as to whether the task has been completed or not.

    Method and system for automating an image rejection process
    14.
    发明授权
    Method and system for automating an image rejection process 有权
    自动化镜像拒绝过程的方法和系统

    公开(公告)号：US09460367B2

    公开(公告)日：2016-10-04

    申请号：US14561512

    申请日：2014-12-05

    Applicant: Xerox Corporation

    Inventor： Vladimir Kozitsky ,  Matthew Shreve ,  Aaron M. Burry

    IPC: G06K9/00 ,  G06K9/46 ,  G06K9/62 ,  G06N99/00

    CPC classification number: G06K9/627 ,  G06K9/00771 ,  G06K9/325 ,  G06K2209/15 ,  G06N99/005 ,  G08G1/0175

    Abstract: Systems and methods for automating an image rejection process. Features including texture, spatial structure, and image quality characteristics can be extracted from one or more images to train a classifier. Features can be calculated with respect to a test image for submission of the features to the classifier, given an operating point corresponding to a desired false positive rate. One or more inputs can be generated from the classifier as a confidence value corresponding to a likelihood of, for example: a license plate being absent in the image, the license plate being unreadable, or the license plate being obstructed. The confidence value can be compared against a threshold to determine if the image(s) should be removed from a human review pipeline, thereby reducing images requiring human review.

    Abstract translation: 用于自动化镜像抑制过程的系统和方法。 可以从一个或多个图像中提取包括纹理，空间结构和图像质量特征的特征来训练分类器。 给定对应于所需假阳性率的操作点，可以相对于用于将特征提交给分类器的测试图像来计算特征。 可以从分类器产生一个或多个输入作为对应于例如图像中不存在牌照，牌照不可读或牌照妨碍的可能性的置信度值。 可以将置信度值与阈值进行比较，以确定图像是否应从人类审查管道中移除，从而减少需要人工审查的图像。

    USING MULTIPLE TRAINED MODELS TO REDUCE DATA LABELING EFFORTS
    16.
    发明公开
    USING MULTIPLE TRAINED MODELS TO REDUCE DATA LABELING EFFORTS 审中-公开

    公开(公告)号：US20240281431A1

    公开(公告)日：2024-08-22

    申请号：US18647425

    申请日：2024-04-26

    Applicant: Xerox Corporation

    Inventor： Matthew Shreve ,  Francisco E. Torres ,  Raja Bala ,  Robert R. Price ,  Pei Li

    IPC: G06F16/23 ,  G06N20/00

    CPC classification number: G06F16/2379 ,  G06N20/00

    Abstract: A method of labeling training data includes inputting a plurality of unlabeled input data samples into each of a plurality of pre-trained neural networks and extracting a set of feature embeddings from multiple layer depths of each of the plurality of pre-trained neural networks. The method also includes generating a plurality of clusterings from the set of feature embeddings. The method also includes analyzing, by a processing device, the plurality of clusterings to identify a subset of the plurality of unlabeled input data samples that belong to a same unknown class. The method also includes assigning pseudo-labels to the subset of the plurality of unlabeled input data samples.

Patent Agency Ranking