公开(公告)号：US11551093B2

公开(公告)日：2023-01-10

申请号：US16254406

申请日：2019-01-22

Applicant: Adobe Inc.

Inventor： Zhe Lin ,  Siyuan Qiao ,  Jianming Zhang

IPC: G06K9/00 ,  G06N3/08 ,  G06N3/04 ,  G06K9/62

Abstract: In implementations of resource-aware training for neural network, one or more computing devices of a system implement an architecture optimization module for monitoring parameter utilization while training a neural network. Dead neurons of the neural network are identified as having activation scales less than a threshold. Neurons with activation scales greater than or equal to the threshold are identified as survived neurons. The dead neurons are converted to reborn neurons by adding the dead neurons to layers of the neural network having the survived neurons. The reborn neurons are prevented from connecting to the survived neurons for training the reborn neurons.

公开(公告)号：US20200234128A1

公开(公告)日：2020-07-23

申请号：US16254406

申请日：2019-01-22

Applicant: Adobe Inc.

Inventor： Zhe Lin ,  Siyuan Qiao ,  Jianming Zhang

IPC: G06N3/08 ,  G06N3/04

Abstract: In implementations of resource-aware training for neural network, one or more computing devices of a system implement an architecture optimization module for monitoring parameter utilization while training a neural network. Dead neurons of the neural network are identified as having activation scales less than a threshold. Neurons with activation scales greater than or equal to the threshold are identified as survived neurons. The dead neurons are converted to reborn neurons by adding the dead neurons to layers of the neural network having the survived neurons. The reborn neurons are prevented from connecting to the survived neurons for training the reborn neurons.

公开(公告)号：US20210232927A1

公开(公告)日：2021-07-29

申请号：US16751897

申请日：2020-01-24

Applicant: Adobe Inc.

Inventor： Zhe Lin ,  Yilin Wang ,  Siyuan Qiao ,  Jianming Zhang

IPC: G06N3/08 ,  G06N20/00

Abstract: In some embodiments, an application receives a request to execute a convolutional neural network model. The application determines the computational complexity requirement for the neural network based on the computing resource available on the device. The application further determines the architecture of the convolutional neural network model by determining the locations of down-sampling layers within the convolutional neural network model based on the computational complexity requirement. The application reconfigures the architecture of the convolutional neural network model by moving the down-sampling layers to the determined locations and executes the convolutional neural network model to generate output results.

公开(公告)号：US20210073644A1

公开(公告)日：2021-03-11

申请号：US16563226

申请日：2019-09-06

Applicant: Adobe Inc.

Inventor： Zhe Lin ,  Yilin Wang ,  Siyuan Qiao ,  Jianming Zhang

IPC: G06N3/08 ,  G06N3/04 ,  G06F7/483

Abstract: A machine learning model compression system and related techniques are described herein. The machine learning model compression system can intelligently remove certain parameters of a machine learning model, without introducing a loss in performance of the machine learning model. Various parameters of a machine learning model can be removed during compression of the machine learning model, such as one or more channels of a single-branch or multi-branch neural network, one or more branches of a multi-branch neural network, certain weights of a channel of a single-branch or multi-branch neural network, and/or other parameters. In some cases, compression is performed only on certain selected layers or branches of the machine learning model. Candidate filters from the selected layers or branches can be removed from the machine learning model in a way that preserves local features of the machine learning model.

公开(公告)号：US12079725B2

公开(公告)日：2024-09-03

申请号：US16751897

申请日：2020-01-24

Applicant: Adobe Inc.

Inventor： Zhe Lin ,  Yilin Wang ,  Siyuan Qiao ,  Jianming Zhang

IPC: G06N20/00 ,  G06N3/082

CPC classification number: G06N3/082 ,  G06N20/00

Abstract: In some embodiments, an application receives a request to execute a convolutional neural network model. The application determines the computational complexity requirement for the neural network based on the computing resource available on the device. The application further determines the architecture of the convolutional neural network model by determining the locations of down-sampling layers within the convolutional neural network model based on the computational complexity requirement. The application reconfigures the architecture of the convolutional neural network model by moving the down-sampling layers to the determined locations and executes the convolutional neural network model to generate output results.

公开(公告)号：US11790234B2

公开(公告)日：2023-10-17

申请号：US18063851

申请日：2022-12-09

Applicant: Adobe Inc.

Inventor： Zhe Lin ,  Siyuan Qiao ,  Jianming Zhang

IPC: G06K9/00 ,  G06N3/082 ,  G06N3/04 ,  G06F18/214

CPC classification number: G06N3/082 ,  G06N3/04 ,  G06F18/2148

Abstract: In implementations of resource-aware training for neural network, one or more computing devices of a system implement an architecture optimization module for monitoring parameter utilization while training a neural network. Dead neurons of the neural network are identified as having activation scales less than a threshold. Neurons with activation scales greater than or equal to the threshold are identified as survived neurons. The dead neurons are converted to reborn neurons by adding the dead neurons to layers of the neural network having the survived neurons. The reborn neurons are prevented from connecting to the survived neurons for training the reborn neurons.