公开(公告)号：US11120364B1

公开(公告)日：2021-09-14

申请号：US16008903

申请日：2018-06-14

Applicant: Amazon Technologies, Inc.

Inventor： Sedat Gokalp ,  Tarun Gupta ,  Abhishek Dan

IPC: G06N20/00 ,  G06N5/04

Abstract: At an artificial intelligence system, respective status indicators for classifier training iterations are determined. A visualization data set comprising the status indicators is presented via an interactive programmatic interface. A training enhancement action, based at least partly on an objective associated with a status indicator, is initiated. The action includes selecting data items for which labeling feedback is to be obtained programmatically during one or more classifier training iterations. A classification model that is trained using the labeling feedback obtained as a result of the action is stored.

公开(公告)号：US20240127575A1

公开(公告)日：2024-04-18

申请号：US18399005

申请日：2023-12-28

Applicant: Amazon Technologies, Inc.

Inventor： Sedat Gokalp ,  Tarun Gupta

IPC: G06V10/75 ,  G06F17/18 ,  G06F18/2113 ,  G06F18/214 ,  G06F18/2411 ,  G06N20/00

CPC classification number: G06V10/758 ,  G06F17/18 ,  G06F18/2113 ,  G06F18/2155 ,  G06F18/2411 ,  G06N20/00

Abstract: Learning iterations, individual ones of which include a respective bucket group selection phase and a class boundary refinement phase, are performed using a source data set whose records are divided into buckets. In the bucket group selection phase of an iteration, a bucket is selected for annotation based on output obtained from a classification model trained in the class boundary refinement phase of an earlier iteration. In the class boundary refinement phase, records of buckets annotated as positive-match buckets for a target class in the bucket group selection phase are selected for inclusion in a training set for a new version of the model using a model enhancement criterion. The trained version of the model is stored.

公开(公告)号：US11893772B1

公开(公告)日：2024-02-06

申请号：US16706472

申请日：2019-12-06

Applicant: Amazon Technologies, Inc.

Inventor： Sedat Gokalp ,  Tarun Gupta

IPC: G06F17/18 ,  G06V10/75 ,  G06N20/00 ,  G06F18/2113 ,  G06F18/214 ,  G06F18/2411

CPC classification number: G06V10/758 ,  G06F17/18 ,  G06F18/2113 ,  G06F18/2155 ,  G06F18/2411 ,  G06N20/00

Abstract: Learning iterations, individual ones of which include a respective bucket group selection phase and a class boundary refinement phase, are performed using a source data set whose records are divided into buckets. In the bucket group selection phase of an iteration, a bucket is selected for annotation based on output obtained from a classification model trained in the class boundary refinement phase of an earlier iteration. In the class boundary refinement phase, records of buckets annotated as positive-match buckets for a target class in the bucket group selection phase are selected for inclusion in a training set for a new version of the model using a model enhancement criterion. The trained version of the model is stored.

公开(公告)号：US11617008B1

公开(公告)日：2023-03-28

申请号：US17218009

申请日：2021-03-30

Applicant: Amazon Technologies, Inc.

Inventor： Tarun Gupta ,  Mayank Sharma ,  Xiang Hao ,  Muhammad Raffay Hamid ,  Zhitao Qiu

IPC: H04N21/439 ,  G06N20/00 ,  H04N21/466 ,  H04N21/475

Abstract: Methods, systems, and computer-readable media for media classification using local and global audio features are disclosed. A media classification system determines local features of an audio input using an audio event detector model that is trained to detect a plurality of audio event classes descriptive of objectionable content. The local features are extracted using maximum values of audio event scores for individual audio event classes. The media classification system determines one or more global features of the audio input using the audio event detector model. The global feature(s) are extracted using averaging of clip-level descriptors of a plurality of clips of the audio input. The media classification system determines a content-based rating for media comprising the audio input based (at least in part) on the local features of the audio input and based (at least in part) on the global feature(s) of the audio input.