Study of validation methods for augmented data

This paper introduces a study to verify whether the expanded data through various data augmentation methods are valid in terms of accuracy and bias. Data augmentation is a method of processing and generating other types of data with similar characteristics based on the characteristics of the obtaine...

Full description

Saved in:
Bibliographic Details
Main Authors: Jung, Jong-Jin, Kim, Kyung-Won
Format: Conference Proceeding
Language:English
Subjects:
biased distribution
data augmentation
embedding vector
Learning (artificial intelligence)
validation
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper introduces a study to verify whether the expanded data through various data augmentation methods are valid in terms of accuracy and bias. Data augmentation is a method of processing and generating other types of data with similar characteristics based on the characteristics of the obtained data, rather than directly collecting data when there is not enough data to increase analysis accuracy. However, unverified and augmented data may actually degrade the results of the analysis. Before using the amplified data for analysis, it is a very important verification factor whether it is accurately propagated in terms of similarity to the source data, and whether bias occurs because only a specific part is concentrated and propagated as a result of the propagation. Therefore, in this paper, a verification method is presented from these two perspectives.
ISSN:2831-6983
DOI:10.1109/ICAIIC57133.2023.10067113