RODA-OOD: Robust Domain Adaptation from Out-of-Distribution Data

Domain adaptation aims to effectively learn from two domains with different distributions, solving labeling problems; however, traditional methods assume that the source and target data are in-distribution data that share the same labels. In practice, Out-Of-Distribution (OOD) data which do not shar...

Full description

Saved in:
Bibliographic Details
Main Authors: Jaekyun Jeong, Mangyu Lee, Sunguk Yun, Keejun Han, Jungeun Kim
Format: Article
Language:English
Published: MDPI AG 2024-12-01
Series:Mathematics
Subjects:
Online Access:https://www.mdpi.com/2227-7390/12/24/3895
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850037364545552384
author Jaekyun Jeong
Mangyu Lee
Sunguk Yun
Keejun Han
Jungeun Kim
author_facet Jaekyun Jeong
Mangyu Lee
Sunguk Yun
Keejun Han
Jungeun Kim
author_sort Jaekyun Jeong
collection DOAJ
description Domain adaptation aims to effectively learn from two domains with different distributions, solving labeling problems; however, traditional methods assume that the source and target data are in-distribution data that share the same labels. In practice, Out-Of-Distribution (OOD) data which do not share labels with the existing data may also be collected during the target data collection process. These OOD data introduce noise and confusion, leading to decreased performance during adaptation. To address this issue, we propose RObust Domain Adaptation from Out-Of-Distribution data (RODA-OOD), a novel method based on data-centric AI principles that focuses on improving data quality rather than refining model architecture. RODA-OOD utilizes the characteristics of deep learning models that prioritize learning in-distribution data, which are easier to train on compared to OOD data. By dynamically adjusting the threshold for OOD detection, the proposed method effectively filters out OOD data, allowing the model to focus on relevant target data. RODA-OOD was compared with competitor and original domain adaptation algorithms based on target data accuracy. The results show that RODA-OOD demonstrates the most robust performance against OOD data, achieving a 21.3% increase in accuracy compared to existing domain adaptation methods. Thus, RODA-OOD can provide a solution to the OOD issue in unsupervised domain adaptation.
format Article
id doaj-art-c1f7ccac25874bdebc0480f293b10fd7
institution DOAJ
issn 2227-7390
language English
publishDate 2024-12-01
publisher MDPI AG
record_format Article
series Mathematics
spelling doaj-art-c1f7ccac25874bdebc0480f293b10fd72025-08-20T02:56:52ZengMDPI AGMathematics2227-73902024-12-011224389510.3390/math12243895RODA-OOD: Robust Domain Adaptation from Out-of-Distribution DataJaekyun Jeong0Mangyu Lee1Sunguk Yun2Keejun Han3Jungeun Kim4Department of Computer Science and Engineering, Kongju National University, Seongbuk-gu, Cheonan 31080, Republic of KoreaDepartment of Computer Science and Engineering, Kongju National University, Seongbuk-gu, Cheonan 31080, Republic of KoreaDepartment of Computer Science and Engineering, Kongju National University, Seongbuk-gu, Cheonan 31080, Republic of KoreaSchool of Computer Engineering, Hansung University, Seongbuk-gu, Seoul 02876, Republic of KoreaDepartment of Computer Engineering, Inha University, Michuhol-gu, Incheon 22212, Republic of KoreaDomain adaptation aims to effectively learn from two domains with different distributions, solving labeling problems; however, traditional methods assume that the source and target data are in-distribution data that share the same labels. In practice, Out-Of-Distribution (OOD) data which do not share labels with the existing data may also be collected during the target data collection process. These OOD data introduce noise and confusion, leading to decreased performance during adaptation. To address this issue, we propose RObust Domain Adaptation from Out-Of-Distribution data (RODA-OOD), a novel method based on data-centric AI principles that focuses on improving data quality rather than refining model architecture. RODA-OOD utilizes the characteristics of deep learning models that prioritize learning in-distribution data, which are easier to train on compared to OOD data. By dynamically adjusting the threshold for OOD detection, the proposed method effectively filters out OOD data, allowing the model to focus on relevant target data. RODA-OOD was compared with competitor and original domain adaptation algorithms based on target data accuracy. The results show that RODA-OOD demonstrates the most robust performance against OOD data, achieving a 21.3% increase in accuracy compared to existing domain adaptation methods. Thus, RODA-OOD can provide a solution to the OOD issue in unsupervised domain adaptation.https://www.mdpi.com/2227-7390/12/24/3895robust domain adaptationout-of-distribution datatemperature scalingcurriculum learningRODA-OOD
spellingShingle Jaekyun Jeong
Mangyu Lee
Sunguk Yun
Keejun Han
Jungeun Kim
RODA-OOD: Robust Domain Adaptation from Out-of-Distribution Data
Mathematics
robust domain adaptation
out-of-distribution data
temperature scaling
curriculum learning
RODA-OOD
title RODA-OOD: Robust Domain Adaptation from Out-of-Distribution Data
title_full RODA-OOD: Robust Domain Adaptation from Out-of-Distribution Data
title_fullStr RODA-OOD: Robust Domain Adaptation from Out-of-Distribution Data
title_full_unstemmed RODA-OOD: Robust Domain Adaptation from Out-of-Distribution Data
title_short RODA-OOD: Robust Domain Adaptation from Out-of-Distribution Data
title_sort roda ood robust domain adaptation from out of distribution data
topic robust domain adaptation
out-of-distribution data
temperature scaling
curriculum learning
RODA-OOD
url https://www.mdpi.com/2227-7390/12/24/3895
work_keys_str_mv AT jaekyunjeong rodaoodrobustdomainadaptationfromoutofdistributiondata
AT mangyulee rodaoodrobustdomainadaptationfromoutofdistributiondata
AT sungukyun rodaoodrobustdomainadaptationfromoutofdistributiondata
AT keejunhan rodaoodrobustdomainadaptationfromoutofdistributiondata
AT jungeunkim rodaoodrobustdomainadaptationfromoutofdistributiondata