Time series domain adaptation via contrastive adversarial domain disentangled network
Unsupervised domain adaptation is a machine learning framework to transform information learned from one or several source domains with many annotated samples to unlabeled target domains. A typical unsupervised domain adaptation method is typically designed base on visual data. Solutions on time se...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Thesis-Master by Research |
Language: | English |
Published: |
Nanyang Technological University
2023
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/168752 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
id |
sg-ntu-dr.10356-168752 |
---|---|
record_format |
dspace |
spelling |
sg-ntu-dr.10356-1687522023-07-04T01:52:13Z Time series domain adaptation via contrastive adversarial domain disentangled network Huang, Xinyi Sinno Jialin Pan School of Computer Science and Engineering sinnopan@ntu.edu.sg Engineering::Computer science and engineering Unsupervised domain adaptation is a machine learning framework to transform information learned from one or several source domains with many annotated samples to unlabeled target domains. A typical unsupervised domain adaptation method is typically designed base on visual data. Solutions on time series data are less explored. Most existing methods are based either on mapping representations from one domain to the other or learning the features invariant to the domains by statistical restrictions. However, focusing only on the mapping or invariant features may leave the methods vulnerable to noise since they ignore much individual information. In this work, we propose Contrastive Adversarial Domain Disentangled Network (CADDN), a novel method that explores improving the model adaptation performance on time series data by exploiting each domain’s specific properties. Our primary motivation is to construct a framework that can learn while jointly disentangling the domain-invariant and the domain-specific features in the mean time. Contrastive learning is applied in the optimization of the domain-specific features, targeting a beneficial and stable feature extraction. Comprehensive experimental evaluations are conducted on four benchmark time series datasets to demonstrate the superiority of the proposed method over state-of-the-art domain adaptation solutions. A further ablation study validates the hypothesis that adding contrastive based domain-specific feature extraction will largely improve the performance compared to only focusing on domain-invariant knowledge. Master of Engineering 2023-06-19T04:30:30Z 2023-06-19T04:30:30Z 2023 Thesis-Master by Research Huang, X. (2023). Time series domain adaptation via contrastive adversarial domain disentangled network. Master's thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/168752 https://hdl.handle.net/10356/168752 10.32657/10356/168752 en This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). application/pdf Nanyang Technological University |
institution |
Nanyang Technological University |
building |
NTU Library |
continent |
Asia |
country |
Singapore Singapore |
content_provider |
NTU Library |
collection |
DR-NTU |
language |
English |
topic |
Engineering::Computer science and engineering |
spellingShingle |
Engineering::Computer science and engineering Huang, Xinyi Time series domain adaptation via contrastive adversarial domain disentangled network |
description |
Unsupervised domain adaptation is a machine learning framework to transform information learned from one or several source domains with many annotated samples to
unlabeled target domains. A typical unsupervised domain adaptation method is typically designed base on visual data. Solutions on time series data are less explored.
Most existing methods are based either on mapping representations from one domain
to the other or learning the features invariant to the domains by statistical restrictions.
However, focusing only on the mapping or invariant features may leave the methods vulnerable to noise since they ignore much individual information. In this work, we propose
Contrastive Adversarial Domain Disentangled Network (CADDN), a novel method that
explores improving the model adaptation performance on time series data by exploiting
each domain’s specific properties. Our primary motivation is to construct a framework
that can learn while jointly disentangling the domain-invariant and the domain-specific
features in the mean time. Contrastive learning is applied in the optimization of the
domain-specific features, targeting a beneficial and stable feature extraction. Comprehensive experimental evaluations are conducted on four benchmark time series datasets to
demonstrate the superiority of the proposed method over state-of-the-art domain adaptation solutions. A further ablation study validates the hypothesis that adding contrastive based domain-specific feature extraction will largely improve the performance compared
to only focusing on domain-invariant knowledge. |
author2 |
Sinno Jialin Pan |
author_facet |
Sinno Jialin Pan Huang, Xinyi |
format |
Thesis-Master by Research |
author |
Huang, Xinyi |
author_sort |
Huang, Xinyi |
title |
Time series domain adaptation via contrastive adversarial domain disentangled network |
title_short |
Time series domain adaptation via contrastive adversarial domain disentangled network |
title_full |
Time series domain adaptation via contrastive adversarial domain disentangled network |
title_fullStr |
Time series domain adaptation via contrastive adversarial domain disentangled network |
title_full_unstemmed |
Time series domain adaptation via contrastive adversarial domain disentangled network |
title_sort |
time series domain adaptation via contrastive adversarial domain disentangled network |
publisher |
Nanyang Technological University |
publishDate |
2023 |
url |
https://hdl.handle.net/10356/168752 |
_version_ |
1772825112465113088 |