The Method for Generating Various Type of Data Using a Single Model and System Thereof

The method and system convert input data into hex codes and time-sequences, using a transformer decoder and distributor to train a model for diverse data generation, addressing limitations of conventional models and enhancing deployment flexibility.

KR103047017B1Active Publication Date: 2026-11-10INDUSTRY UNIVERSITY COOPERATION FOUNDATION HANYANG UNIVERSITY
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
KR1020230087411
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2023-07-05
Publication Date
2026-11-10
Estimated Expiration
2043-07-05

AI Technical Summary

Technical Problem

Conventional generative AI models are limited to generating specific types of data and require data size matching during preprocessing or postprocessing, which can negatively impact data quality.

Method used

A data generation method and system that converts various types of data into hex codes and time-sequences, performs masking preprocessing based on batch size, and trains a model using a transformer decoder and distributor to generate diverse data types.

Benefits of technology

Enables a single model to generate various types of data efficiently, reducing memory requirements and allowing deployment across different devices, from large-scale servers to wearable technology.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112023074161390-PAT00002_ABST
    Figure 112023074161390-PAT00002_ABST
Patent Text Reader

Abstract

A data generation method according to a disclosed embodiment comprises: a step of processing various types of input data collected from the outside; a step of training an artificial intelligence model to generate time-series output data by inputting a class token including the processed input data and type information of data to be generated; and a step of generating output data based on type information from the trained artificial intelligence model when a user inputs type information to be generated; wherein the step of processing the input data comprises coding the input data based on a preset reference value, time-series the coded input data, and performing masking processing on the time-series input data based on the batch size of the artificial intelligence model.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Computing device and related product

    CN109583579A

  • Fault diagnosis method for lightweight electromechanical equipment

    CN114722923A

  • Electronic apparatus and controlling method thereof

    US20220215034A1

  • Systems and methods for implementing data transformations in multiple execution contexts

    US20220269978A1