公开(公告)号：US20210174210A1

公开(公告)日：2021-06-10

申请号：US16729480

申请日：2019-12-30

Applicant: Industrial Technology Research Institute

Inventor： Ming-Chun Hsyu ,  Chao-Hong Chen ,  Chien-Chih Huang

IPC: G06N3/08 ,  G06N3/04 ,  G06N20/00 ,  G06K9/62

Abstract: A method and an electronic device for selecting deep neural network hyperparameters are provided. In an embodiment of the method, a plurality of testing hyperparameter configurations are sampled from a plurality of hyperparameter ranges of a plurality of hyperparameters. A target neural network model is trained by using a training dataset and the plurality of testing hyperparameter configurations, and a plurality of accuracies corresponding to the plurality of testing hyperparameter configurations are obtained after training for preset epochs. A hyperparameter recommendation operation is performed to predict a plurality of final accuracies of the plurality of testing hyperparameter configurations. A recommended hyperparameter configuration corresponding to the final accuracy having a highest predicted value is selected as a hyperparameter setting for continuing training the target neural network model.

公开(公告)号：US11537893B2

公开(公告)日：2022-12-27

申请号：US16729480

申请日：2019-12-30

Applicant: Industrial Technology Research Institute

Inventor： Ming-Chun Hsyu ,  Chao-Hong Chen ,  Chien-Chih Huang

IPC: G06N3/08 ,  G06N3/04 ,  G06N20/00 ,  G06K9/62

Abstract: A method and an electronic device for selecting deep neural network hyperparameters are provided. In an embodiment of the method, a plurality of testing hyperparameter configurations are sampled from a plurality of hyperparameter ranges of a plurality of hyperparameters. A target neural network model is trained by using a training dataset and the plurality of testing hyperparameter configurations, and a plurality of accuracies corresponding to the plurality of testing hyperparameter configurations are obtained after training for preset epochs. A hyperparameter recommendation operation is performed to predict a plurality of final accuracies of the plurality of testing hyperparameter configurations. A recommended hyperparameter configuration corresponding to the final accuracy having a highest predicted value is selected as a hyperparameter setting for continuing training the target neural network model.

公开(公告)号：US12299565B2

公开(公告)日：2025-05-13

申请号：US17135793

申请日：2020-12-28

Applicant: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

Inventor： Chao-Hung Chen ,  Ming-Chun Hsyu ,  Chien-Chih Huang ,  Wen-Pin Hsu ,  Chun-Te Yu

IPC: G06N3/08 ,  G06F7/50 ,  G06F7/523 ,  G06F18/21

Abstract: A data processing method used in neural network computing is provided. During a training phase of a neural network model, a feedforward procedure based on a calibration data is performed to obtain distribution information of a feedforward result for at least one layer of the neural network model. During the training phase of the neural network model, a bit upper bound of a partial sum is generated based on the distribution information of the feedforward result. During an inference phase of the neural network model, a bit-number reducing process is performed on an original operation result of an input data and a weight for the neural network model according to the bit upper bound of the partial sum to obtain an adjusted operation result.