公开(公告)号：US20240012873A1

公开(公告)日：2024-01-11

申请号：US18077126

申请日：2022-12-07

Applicant: Industrial Technology Research Institute

Inventor： Chia-Hsiang Yang ,  Chen-Chien Kao ,  Chao-Hung Chen

IPC: G06F17/16 ,  G06F17/14

CPC classification number: G06F17/16 ,  G06F17/145 ,  G06F17/147

Abstract: An electronic device and a method for accelerating canonical polyadic (CP) decomposition are provided. The method includes: performing at least one of a Walsh-Hadamard transform (WHT) operation and a discrete cosine transform (DCT) operation on a first factor matrix, a second factor matrix, and a tensor respectively to update the first factor matrix, the second factor matrix and the tensor; sampling the updated first factor matrix and the updated second factor matrix to generate a first sampled matrix, and sampling an unfolded matrix of the updated tensor to generate a second sampled matrix; solving a least square problem of the first sampled matrix and the second sampled matrix to generate or update a third factor matrix of the tensor so as to update multiple components of the tensor; and outputting multiple components after an updating of multiple components is finished.

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.