公开(公告)号：US20230196094A1

公开(公告)日：2023-06-22

申请号：US17560010

申请日：2021-12-22

Applicant: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

Inventor： Chih-Cheng LU ,  Jin-Yu LIN ,  Kai-Cheung JUANG

IPC: G06N3/08 ,  G06N3/04

CPC classification number: G06N3/08 ,  G06N3/0481

Abstract: A quantization method for neural network model includes following steps: initializing a weight array of a neural network model, wherein the weight array includes a plurality of initial weights; performing a quantization procedure to generate a quantized weight array according to the weight array, wherein the quantized weight array includes a plurality of quantized weights within a fixed range; performing a training procedure of the neural network model according to the quantized weight array; and determining whether a loss function is convergent in the training procedure and outputting a post-trained quantized weight array when the loss function is convergent.

公开(公告)号：US20210199503A1

公开(公告)日：2021-07-01

申请号：US16727382

申请日：2019-12-26

Applicant: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

Inventor： Chih-Cheng LU ,  Kai-Cheung JUANG

IPC: G01J3/28 ,  G06N20/00

Abstract: A data processing system disposed on a sensor comprises a de-identified sensing device and a decoding device. The de-identified sensing device is configured to receive a sensing data of a target and to process the sensing data to generate a de-identified data. The decoding device communicably connects to the de-identified sensing device and is configured to generate a decoded data according to the de-identified data and a decoding parameter obtained from a database trained by machine learning. The de-identified sensing device comprises an analog encoder configured to encode the sensing data to generate a responsive data.

公开(公告)号：US20240194217A1

公开(公告)日：2024-06-13

申请号：US18089189

申请日：2022-12-27

Applicant: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

Inventor： Chih-Cheng LU ,  Jian-Bai LI ,  Cheng-Ming SHIH ,  Yu-Lee YEH ,  Kai-Cheung JUANG

IPC: G10L25/18

CPC classification number: G10L25/18 ,  G10L2025/937

Abstract: A data processing method for acoustic event includes: establishing a simulated acoustic frequency event module, a data capturing module, and a sound application decision module in a software manner, setting a simulated hardware parameter to the simulated acoustic frequency event module, inputting a sound signal to a frequency filtering module of the simulated acoustic frequency event module, and obtaining metadata from a frequency event quantizer of the simulated acoustic frequency event module, dividing each of the metadata into multiple frames according to a time interval by the data capturing module, accumulating an event number of each frame by the data capturing module, setting a label of each frame according to the event number, storing these frames, the event number and the label in a database, and training a decision model by the sound application decision module according to the database and a sound application.